[CVPR 2022] Structured Sparse R-CNN for Direct Scene Graph Generation

Related tags

Deep Learning Structured-Sparse-RCNN
Overview

Structured Sparse R-CNN for Direct Scene Graph Generation

Our paper Structured Sparse R-CNN for Direct Scene Graph Generation has been accepted by CVPR 2022.

Requirements

Environments: python 3.8, cuda 10.1, pytorch 1.7.1

To install requirements:

conda create --name scene_graph_benchmark
conda activate scene_graph_benchmark

pip install --user ipython
pip install --user scipy
pip install --user h5py
pip install --user pyyaml
pip install --user yacs
pip install --user scipy
pip install --user h5py
pip install --user tqdm
pip install --user opencv-python

pip install --user ninja yacs cython matplotlib tqdm opencv-python overrides


conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=10.1 -c pytorch

git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI
python setup.py build_ext install

git clone https://github.com/NVIDIA/apex.git
cd apex
python setup.py install --cuda_ext --cpp_ext

# put our code SGGbench into your directory, such as /home/username
cd /home/username/SGGbench 
python setup.py build develop

Datasets

VG

For the dataset and pretrained backbone weights preparation, please follow: Scene-Graph-Benchmark.pytorch

OI v4, v6

For the datasets and pretrained backbone weights preparation, please follow: BGNN-SGG. Actually, BGNN-SGG is also compatible with Scene-Graph-Benchmark.pytorch.

Training

VG

On VG, we notice ~0.2 R@100 noise for our model with 300 queries.

To train the model with 300 triplet queries in the paper, run this command. The results are R@100: 36.9; zR@100: 3.7; mR@100: 10.0: (8 RTX 2080ti 11G)

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_train_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/pretrained_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

To train the model with 800 triplet queries in the paper, run this command. The results are R@100: 38.4; zR@100: 4.0; mR@100: 10.3: (8 V100 32G)

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_train_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/pretrained_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 800 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

To train the model with 300 triplet queries and backbone, run this command. The results are R@100: 36.7; zR@100: 3.8; mR@100: 10.1: (8 V100 32G)

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_train_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" SOLVER.BACKBONE_MULTIPLIER 0.1 MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/pretrained_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel_bkb MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE False MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

OI V4

To train the model with 300 triplet queries in the paper, run this command: (8 RTX 2080ti 11G)

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_train_net.py --config-file "configs/oiv4_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/iov4_pretrain_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

OI V6

To train the model with 300 triplet queries in the paper, run this command: (8 RTX 2080ti 11G)

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_train_net.py --config-file "configs/oiv6_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/iov6_pretrain_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

Evaluation

VG

300 queries

To evaluate my model, run:

CUDA_VISIBLE_DEVICES=4,5,6,7 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/eva_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'eva_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with TDE, run:

CUDA_VISIBLE_DEVICES=2,3,4,5 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/tde_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS True MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'tde_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with our LA, run:

CUDA_VISIBLE_DEVICES=4,5,6,7 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/la03_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True MODEL.SimrelRCNN.REL_LOGITS_ADJUSTMENT True MODEL.SimrelRCNN.LOGIT_ADJ_TAU 0.3

📋 Before testing, create a new directory, named 'la03_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

800 queries

To evaluate my model, run:

CUDA_VISIBLE_DEVICES=3,5,6,7 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/eva_real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 800 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'eva_real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with TDE, run:

CUDA_VISIBLE_DEVICES=2,3,4,5 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/tde_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 800 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS True MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'tde_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with our LA, run:

CUDA_VISIBLE_DEVICES=3,5,6,7 python -m torch.distributed.launch --master_port 10029 --nproc_per_node=4 tools/relation_test_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 4 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/LA03_real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 800 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True MODEL.SimrelRCNN.REL_LOGITS_ADJUSTMENT True MODEL.SimrelRCNN.LOGIT_ADJ_TAU 0.3

📋 Before testing, create a new directory, named 'LA03_real025kl_newe2rposition_fullobjbranch_800q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

OI V4

300 queries

To evaluate my model, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_test_net.py --config-file "configs/oiv4_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/eva_oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'eva_oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with our LA, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_test_net.py --config-file "configs/oiv4_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/la_oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True MODEL.SimrelRCNN.REL_LOGITS_ADJUSTMENT True MODEL.SimrelRCNN.LOGIT_ADJ_TAU 0.3

📋 Before testing, create a new directory, named 'la_oiv4_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

OI V6

300 queries

To evaluate my model, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_test_net.py --config-file "configs/oiv6_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/eva_oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

📋 Before testing, create a new directory, named 'eva_oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

To evaluate my model with our LA, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=8 tools/relation_test_net.py --config-file "configs/oiv6_simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.WEIGHT Outputs/oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel/model_final.pth OUTPUT_DIR Outputs/la_oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 1024 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True MODEL.SimrelRCNN.REL_LOGITS_ADJUSTMENT True MODEL.SimrelRCNN.LOGIT_ADJ_TAU 0.3

📋 Before testing, create a new directory, named 'la_oiv6_real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel', in ./SGGbench/Outputs/ .

Pre-trained Models

You can download pretrained models here:

VG

300 queries

800 queries

OI V4

300 queries

OI V6

300 queries

Results

Our model achieves the following performance on :

Visual Genome

* means 800 queries.

Models SGGen R@20 SGGen R@50 SGGen R@100 SGGen zR@20 SGGen zR@50 SGGen zR@100 SGGen mR@20 SGGen mR@50 SGGen mR@100
Our model 25.8 32.7 36.9 1.5 2.7 3.7 6.1 8.4 10.0
Our model* 26.1 33.5 38.4 1.5 2.7 4.0 6.2 8.6 10.3
Our model+TDE 14.5 18.3 21.0 1.8 2.7 3.6 10.8 15.0 18.5
Our model*+TDE 15.0 19.7 22.9 1.6 2.7 3.8 9.8 14.6 18.0
Our model+LA 18.4 23.3 26.5 1.9 2.9 4.0 13.5 17.9 21.4
Our model*+LA 18.2 23.7 27.3 2.0 3.1 4.5 13.7 18.6 22.5

Acknowledgement

Our code is mainly based on: Scene-Graph-Benchmark.pytorch, SparseR-CNN and BGNN-SGG.

For this paper, I'm extremely grateful to my advisor Prof. Limin Wang. I should also appreciate my group members for discussing with me: Jing Tan, Ziteng Gao and Jiaqi Tang.

Citations

@inproceedings{ssrcnnsgg22cvpr,
  author    = {Yao Teng and
               Limin Wang},
  title     = {Structured Sparse {R-CNN} for Direct Scene Graph Generation},
  booktitle = {{CVPR}},
  year      = {2022}
}
Comments
  • problem of reproducing training process

    problem of reproducing training process

    What I expect:

    Train the model from scratch as instructed in readme.

    What I did:

    I prepared the dataset as instructed in Scene-Graph-Benchmark.pytorch, including preparing images and the labels.

    I run the following command as instructed in readme (without distributed mode because I only use one GPU):

    python tools/relation_train_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/pretrained_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

    What the problem is:

    The log first shows numerous cuda errors like:

    /opt/conda/conda-bld/pytorch_1603729047590/work/aten/src/ATen/native/cuda/IndexKernel.cu:84: operator(): block: [46,0,0], thread: [127,0,0] Assertion index >= -sizes[i] && index < sizes[i] && "index out of bounds" failed.

    Then some python errors:

    Traceback (most recent call last):

    File "/home/visionx/software/anaconda2/envs/pysgg2/lib/python3.7/site-packages/torch/autograd/grad_mode.py", line 26, in decorate_context

    return func(*args, **kwargs)

    File "/home/visionx/data6T/xuezhi/Structured-Sparse-RCNN/maskrcnn_benchmark/modeling/roi_heads/simrel_head/loss_simrel.py", line 1156, in forward

    cost_class_s = cost_class_s[:, X]

    RuntimeError: CUDA error: device-side assert triggered

    I observed that 'cost_class_s' is of dim (..., 100), the values in 'X' range from 0 to 99 normally, but for a certain instance, the range becomes 100 to 199.

    opened by wishforgood 18
  • Some questions about training

    Some questions about training

    Hi, thanks for your nice work !

    When I tried to reproduce the results (4*3090(24G), 300queries, batch=8) with the commends

    CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --master_port 10020 --nproc_per_node=4 tools/relation_train_net.py --config-file "configs/simrel_e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR CausalAnalysisPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE none MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE sum MODEL.ROI_RELATION_HEAD.CAUSAL.CONTEXT_LAYER motifs SOLVER.IMS_PER_BATCH 8 TEST.IMS_PER_BATCH 8 SOLVER.VAL_PERIOD 12000 SOLVER.MAX_ITER 80000 SOLVER.STEPS '(47000, 64000)' SOLVER.BASE_LR 0.000008 SOLVER.OPTIMIZER "ADAMW" MODEL.PRETRAINED_DETECTOR_CKPT checkpoints/pretrained_faster_rcnn/model_final.pth OUTPUT_DIR Outputs/real025kl_newe2rposition_fullobjbranch_300q_smalldyconv_prequeryaug9_norel_hardlabel MODEL.ROI_BOX_HEAD.MLP_HEAD_DIM 1024 MODEL.SimrelRCNN.ENABLE_BG_OBJ False MODEL.SimrelRCNN.ENABLE_REL_X2Y True MODEL.SimrelRCNN.REL_DIM 256 MODEL.SimrelRCNN.NUM_PROPOSALS 300 MODEL.SimrelRCNN.NUM_HEADS 6 MODEL.SimrelRCNN.REL_STACK_NUM 6 MODEL.SimrelRCNN.TRIPLET_MASK_WEIGHT 1.0 MODEL.SimrelRCNN.FREEZE_BACKBONE True MODEL.SimrelRCNN.CROSS_OBJ_FEAT_FUSION False MODEL.SimrelRCNN.CLASS_WEIGHT 1.333 MODEL.SimrelRCNN.L1_WEIGHT 5.0 MODEL.SimrelRCNN.GIOU_WEIGHT 2.0 GLOVE_DIR '' MODEL.SimrelRCNN.ENABLE_FREQ False MODEL.SimrelRCNN.ENABLE_QUERY_REVERSE False MODEL.SimrelRCNN.USE_REFINE_OBJ_FEATURE True MODEL.SimrelRCNN.FREEZE_PUREE_OBJDET False MODEL.SimrelRCNN.PURE_ENT_NUM_PROPOSALS 100 MODEL.SimrelRCNN.ENABLE_MASK_BRANCH False MODEL.SimrelRCNN.KL_BRANCH_WEIGHT 0.25 MODEL.SimrelRCNN.ENABLE_KL_BRANCH True MODEL.SimrelRCNN.POSI_ENCODE_DIM 64 MODEL.SimrelRCNN.REL_CLASS_WEIGHT 1.334 MODEL.SimrelRCNN.PURE_ENT_CLASS_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_GIOU_WEIGHT 2.0 MODEL.SimrelRCNN.PURE_ENT_L1_WEIGHT 5.0 MODEL.SimrelRCNN.AUXILIARY_BRANCH True MODEL.SimrelRCNN.AUXILIARY_BRANCH_SELECT_ENT_MAX_NUM 25 MODEL.SimrelRCNN.AUXILIARY_BRANCH_START 11 MODEL.SimrelRCNN.ENABLE_ENT_PROP True MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_ANALYSIS False MODEL.SimrelRCNN.USE_CROSS_RANK False MODEL.SimrelRCNN.DISABLE_KQ_FUSION_SELFATTEN False MODEL.SimrelRCNN.DIM_ENT_PRE_CLS 256 MODEL.SimrelRCNN.DIM_ENT_PRE_REG 256 MODEL.SimrelRCNN.ONE_REL_CONV True MODEL.SimrelRCNN.ENABLE_BATCH_REDUCTION True MODEL.SimrelRCNN.USE_HARD_LABEL_KLMATCH True MODEL.SimrelRCNN.DISABLE_OBJ2REL_LOSS True

    The training process always stop at iter:7200 without any errors. The volatile GPU-util for 4 gpus are all 100% and the memory usages are normal. T^T (But when I use bs=4 (1 sample for every gpu), the problem goes away. ) To fully utilize the four gpus, I think bs=8 is a better choice but I couldn't find a solution to the aforementioned problem. Could you help me figure out that problem? Thanks a lot!

    What I have tried: Change num_workers from 4 to 1 to 0 ×

    opened by Coco-wyc 5
  • error when eval

    error when eval

    File "/root/Structured-Sparse-RCNN/maskrcnn_benchmark/modeling/roi_heads/simrel_head/model_simrel.py", line 1163, in forward pro_max_cls_score, pro_cls_id = torch.max(class_logits, dim=-1) UnboundLocalError: local variable 'class_logits' referenced before assignment

    opened by ZHUXUHAN 3
  • Training scripts and results on predcls & sgcls

    Training scripts and results on predcls & sgcls

    Hi, thanks for your nice work :) Could you provide the training scripts and results on predcls and sgcls? They are two common settings in scene graph generation.

    opened by jkli-aa 2
  • About method: how to tell two objects in two triplet are the same object or not?

    About method: how to tell two objects in two triplet are the same object or not?

    Hi, thanks for sharing your work! I read your paper, and got a question about the formulation of triplet detection: how to tell two objects in two triplets are the same object or not? For example, in Fig.1, "woman wearing shoe" and "woman wearing shirt" are referring to the same "woman", but the triplet detection formulation will predict the "woman" box twice for these 2 triplet. How to tell if these two "woman" boxes are actually the same one? I'm not familiar with SGG task, but it seems this detail is important for evaluation.

    opened by Charles-Xie 2
Owner
Multimedia Computing Group, Nanjing University
Multimedia Computing Group, Nanjing University
GitHub
Pytorch implementation of Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priors

Make-A-Scene - PyTorch Pytorch implementation (inofficial) of Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priors (https://arxiv.org/

Casual GAN Papers 259 Dec 28, 2022
Code for CVPR 2021 oral paper "Exploring Data-Efficient 3D Scene Understanding with Contrastive Scene Contexts"

Exploring Data-Efficient 3D Scene Understanding with Contrastive Scene Contexts The rapid progress in 3D scene understanding has come with growing dem

Facebook Research 182 Dec 30, 2022
Neural Scene Graphs for Dynamic Scene (CVPR 2021)

Implementation of Neural Scene Graphs, that optimizes multiple radiance fields to represent different objects and a static scene background. Learned representations can be rendered with novel object compositions and views.

null 151 Dec 26, 2022
Implementation of CVPR'2022:Reconstructing Surfaces for Sparse Point Clouds with On-Surface Priors

Reconstructing Surfaces for Sparse Point Clouds with On-Surface Priors (CVPR 2022) Personal Web Pages | Paper | Project Page This repository contains

null 151 Dec 26, 2022
Focal Sparse Convolutional Networks for 3D Object Detection (CVPR 2022, Oral)

Focal Sparse Convolutional Networks for 3D Object Detection (CVPR 2022, Oral) This is the official implementation of Focals Conv (CVPR 2022), a new sp

DV Lab 280 Jan 7, 2023
Official PyTorch code of DeepPanoContext: Panoramic 3D Scene Understanding with Holistic Scene Context Graph and Relation-based Optimization (ICCV 2021 Oral).

DeepPanoContext (DPC) [Project Page (with interactive results)][Paper] DeepPanoContext: Panoramic 3D Scene Understanding with Holistic Scene Context G

Cheng Zhang 66 Nov 16, 2022
Differentiable Neural Computers, Sparse Access Memory and Sparse Differentiable Neural Computers, for Pytorch

Differentiable Neural Computers and family, for Pytorch Includes: Differentiable Neural Computers (DNC) Sparse Access Memory (SAM) Sparse Differentiab

ixaxaar 302 Dec 14, 2022
Sparse R-CNN: End-to-End Object Detection with Learnable Proposals, CVPR2021

End-to-End Object Detection with Learnable Proposal, CVPR2021

Peize Sun 1.2k Dec 27, 2022
NFT-Price-Prediction-CNN - Using visual feature extraction, prices of NFTs are predicted via CNN (Alexnet and Resnet) architectures.

NFT-Price-Prediction-CNN - Using visual feature extraction, prices of NFTs are predicted via CNN (Alexnet and Resnet) architectures.

null 5 Nov 3, 2022
HSC4D: Human-centered 4D Scene Capture in Large-scale Indoor-outdoor Space Using Wearable IMUs and LiDAR. CVPR 2022

HSC4D: Human-centered 4D Scene Capture in Large-scale Indoor-outdoor Space Using Wearable IMUs and LiDAR. CVPR 2022 [Project page | Video] Getting sta

null 51 Nov 29, 2022
Pytorch re-implementation of Paper: SwinTextSpotter: Scene Text Spotting via Better Synergy between Text Detection and Text Recognition (CVPR 2022)

SwinTextSpotter This is the pytorch implementation of Paper: SwinTextSpotter: Scene Text Spotting via Better Synergy between Text Detection and Text R

mxin262 183 Jan 3, 2023
Code for "Neural 3D Scene Reconstruction with the Manhattan-world Assumption" CVPR 2022 Oral

News 05/10/2022 To make the comparison on ScanNet easier, we provide all quantitative and qualitative results of baselines here, including COLMAP, COL

ZJU3DV 365 Dec 30, 2022
Code for "Learning Canonical Representations for Scene Graph to Image Generation", Herzig & Bar et al., ECCV2020

Learning Canonical Representations for Scene Graph to Image Generation (ECCV 2020) Roei Herzig*, Amir Bar*, Huijuan Xu, Gal Chechik, Trevor Darrell, A

roei_herzig 24 Jul 7, 2022
image scene graph generation benchmark

Scene Graph Benchmark in PyTorch 1.7 This project is based on maskrcnn-benchmark Highlights Upgrad to pytorch 1.7 Multi-GPU training and inference Bat

Microsoft 303 Dec 27, 2022
Imposter-detector-2022 - HackED 2022 Team 3IQ - 2022 Imposter Detector

HackED 2022 Team 3IQ - 2022 Imposter Detector By Aneeljyot Alagh, Curtis Kan, Jo

Joshua Ji 3 Aug 20, 2022
Semi-supervised Implicit Scene Completion from Sparse LiDAR

Semi-supervised Implicit Scene Completion from Sparse LiDAR Paper Created by Pengfei Li, Yongliang Shi, Tianyu Liu, Hao Zhao, Guyue Zhou and YA-QIN ZH

null 114 Nov 30, 2022
A simple algorithm for extracting tree height in sparse scene from point cloud data.

TREE HEIGHT EXTRACTION IN SPARSE SCENES BASED ON UAV REMOTE SENSING This is the offical python implementation of the paper "Tree Height Extraction in

null 6 Oct 28, 2022
GDR-Net: Geometry-Guided Direct Regression Network for Monocular 6D Object Pose Estimation. (CVPR 2021)

GDR-Net This repo provides the PyTorch implementation of the work: Gu Wang, Fabian Manhardt, Federico Tombari, Xiangyang Ji. GDR-Net: Geometry-Guided

null 169 Jan 7, 2023
The 7th edition of NTIRE: New Trends in Image Restoration and Enhancement workshop will be held on June 2022 in conjunction with CVPR 2022.

NTIRE 2022 - Image Inpainting Challenge Important dates 2022.02.01: Release of train data (input and output images) and validation data (only input) 2

Andrés Romero 37 Nov 27, 2022