PyTorch implemention of ICCV'21 paper SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation

Last update: Dec 5, 2022

Related tags

Overview

SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation

This is the PyTorch implemention of ICCV'21 paper SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation by Kai Chen and Qi Dou.

Abstract

Category-level 6D object pose estimation aims to predict the position and orientation for unseen objects, which plays a pillar role in many scenarios such as robotics and augmented reality. The significant intra-class variation is the bottleneck challenge in this task yet remains unsolved so far. In this paper, we take advantage of category prior to overcome this problem by innovating a structure-guided prior adaptation scheme to accurately estimate 6D pose for individual objects. Different from existing prior based methods, given one object and its corresponding category prior, we propose to leverage their structure similarity to dynamically adapt the prior to the observed object. The prior adaptation intrinsically associates the adopted prior with different objects, from which we can accurately reconstruct the 3D canonical model of the specific object for pose estimation. To further enhance the structure characteristic of objects, we extract low-rank structure points from the dense object point cloud, therefore more efficiently incorporating sparse structural information during prior adaptation. Extensive experiments on CAMERA25 and REAL275 benchmarks demonstrate significant performance improvement.

Requirements

Linux (tested on Ubuntu 18.04)
Python 3.6+
CUDA 10.0
PyTorch 1.1.0

Installation

Conda virtual environment

We recommend using conda to setup the environment.

If you have already installed conda, please use the following commands.

conda create -n sgpa python=3.6
conda activate sgpa
pip install -r requirements.txt

Build PointNet++

cd SGPA/pointnet2/pointnet2
python setup.py install

Build nn_distance

cd SGPA/lib/nn_distance
python setup.py install

Dataset

Download camera_train, camera_val, real_train, real_test, ground-truth annotations and mesh models provided by NOCS.

Then, organize and preprocess these files following SPD. For a quick evaluation, we provide the processed testing data for REAL275. You can download it here and organize the testing data as follows:

SGPA
├── data
│   └── Real
│       ├──test
│       └──test_list.txt
└── results
    └── mrcnn_results
        └──real_test

Evaluation

Please download our trained model here and put it in the 'SGPA/model' directory. Then, you can have a quick evaluation on the REAL275 dataset using the following command.

bash eval.sh

Train

In order to train the model, remember to download the complete dataset, organize and preprocess the dataset properly at first.

train.py is the main file for training. You can simply start training using the following command.

bash train.sh

Citation

If you find the code useful, please cite our paper.

@inproceedings{chen2021sgpa,
  title={Sgpa: Structure-guided prior adaptation for category-level 6d object pose estimation},
  author={Chen, Kai and Dou, Qi},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={2773--2782},
  year={2021}
}

Any questions, please feel free to contact Kai Chen ([email protected]).

Acknowledgment

The dataset is provided by NOCS. Our code is developed based on SPD and Pointnet2.PyTorch.

Comments

Code to generate prior point clouds

Hi there,

first of all, thank you for your fascinating work!

In the paper I read this part:

For the prior point cloud, we train an autoencoder network on the ShapeNet dataset [3], and then we feed the average embedding of all instances that belong to the same category into the trained decoder to get the prior point cloud for this category.

I couldn't find code for this step in the repository. Did I miss something? Would it be possible to upload this part too? Reason being that I'd like to train with further object categories.

Thanks!

opened by kannwism 2
Cannot reproduce the results in the paper

Hi, thank you for open source this work, I retrained your model without changing the code, but the experimental results are quite different (some metrics differ by nearly 10 percent), I would like to know what caused this, the following are my environment settings: Pytorch 1.7.0 Cuda 11.0.3 A single RTX3090

opened by CNJianLiu 2
what's the performance of SGPA using pointnet rather than pointnet++？

Hi, author. Thank you for releasing such a wonderful work!

I have a question about "the performance of the model when replacing PointNet + + in SGPA with PointNet? Because SPD adopts PointNet instead of PointNet++." It seems that there are no ablation results for this change.

opened by Bingo-1996 2
Question about the value of train_steps

Thanks for sharing good work. I tried to train SGPA on CAMERA+Real. However, it seems that the something wrong in sampling CAMERA dataset when running train.sh. raise ValueError: Sample larger than population or is negative

This error means that the value of 3*n_repeat*real_len is larger than len(camera_indices) data_list = random.sample(camera_indices, 3*n_repeat*real_len) + real_indices*n_repeat in train.py line 100

The output of related variables as follows:

For comparison, I run the corresponding code in SPD, output as follows:

It seems that the difference of train_steps and batch_size makes the value of train_size bigger than len(camera_indices), and the influence of train_steps is the main one(SGPA:8000, SPD:1500) I use dataset preprocess code from SPD, the reason I can think of is that the dataset is different, or the value of train_steps is wrong. Did I miss anything? Hope someone provide the output of the values I listed. :)

opened by winka9587 1

The Degeneration of Assign_Matrix.

I tried to train SGPA on CAMERA25-Composed. However, it seems that it will always drop in a degeneration, where the entropy_loss will drop dramatically to zero. Here is the log:

2022-04-11 07:20:52,242 : Batch 5630 Loss:0.022436, corr_loss:0.015928, cd_loss:0.005587, entropy_loss:0.000209, deform_loss:0.000313, lowrank_loss:0.000399
2022-04-11 07:20:58,618 : Batch 5640 Loss:0.061784, corr_loss:0.045377, cd_loss:0.015464, entropy_loss:0.000174, deform_loss:0.000297, lowrank_loss:0.000472
2022-04-11 07:21:04,251 : Batch 5650 Loss:0.107389, corr_loss:0.086328, cd_loss:0.019944, entropy_loss:0.000153, deform_loss:0.000268, lowrank_loss:0.000695
2022-04-11 07:21:10,448 : Batch 5660 Loss:0.133275, corr_loss:0.105933, cd_loss:0.025799, entropy_loss:0.000137, deform_loss:0.000216, lowrank_loss:0.001190
2022-04-11 07:21:17,287 : Batch 5670 Loss:0.125959, corr_loss:0.103875, cd_loss:0.021085, entropy_loss:0.000160, deform_loss:0.000186, lowrank_loss:0.000652
2022-04-11 07:21:22,624 : Batch 5680 Loss:0.144936, corr_loss:0.126369, cd_loss:0.014371, entropy_loss:0.000087, deform_loss:0.000158, lowrank_loss:0.003951
2022-04-11 07:21:27,534 : Batch 5690 Loss:0.149627, corr_loss:0.127070, cd_loss:0.020821, entropy_loss:0.000024, deform_loss:0.000137, lowrank_loss:0.001574
2022-04-11 07:21:32,971 : Batch 5700 Loss:0.136263, corr_loss:0.121296, cd_loss:0.014015, entropy_loss:0.000021, deform_loss:0.000140, lowrank_loss:0.000791
2022-04-11 07:21:39,646 : Batch 5710 Loss:0.159056, corr_loss:0.132856, cd_loss:0.024632, entropy_loss:0.000019, deform_loss:0.000150, lowrank_loss:0.001400
2022-04-11 07:21:45,321 : Batch 5720 Loss:0.154485, corr_loss:0.128831, cd_loss:0.023682, entropy_loss:0.000012, deform_loss:0.000141, lowrank_loss:0.001818
2022-04-11 07:21:50,838 : Batch 5730 Loss:0.189477, corr_loss:0.158418, cd_loss:0.027742, entropy_loss:0.000011, deform_loss:0.000149, lowrank_loss:0.003158
2022-04-11 07:21:57,104 : Batch 5740 Loss:0.123445, corr_loss:0.107106, cd_loss:0.013482, entropy_loss:0.000011, deform_loss:0.000129, lowrank_loss:0.002717
2022-04-11 07:22:03,384 : Batch 5750 Loss:0.133284, corr_loss:0.111759, cd_loss:0.019112, entropy_loss:0.000010, deform_loss:0.000132, lowrank_loss:0.002271
2022-04-11 07:22:08,641 : Batch 5760 Loss:0.123830, corr_loss:0.107223, cd_loss:0.013527, entropy_loss:0.000012, deform_loss:0.000139, lowrank_loss:0.002929
2022-04-11 07:22:14,902 : Batch 5770 Loss:0.282613, corr_loss:0.088271, cd_loss:0.015968, entropy_loss:0.000015, deform_loss:0.000141, lowrank_loss:0.178218
2022-04-11 07:22:19,834 : Batch 5780 Loss:0.140194, corr_loss:0.119629, cd_loss:0.018370, entropy_loss:0.000006, deform_loss:0.000141, lowrank_loss:0.002047
2022-04-11 07:22:26,458 : Batch 5790 Loss:0.147937, corr_loss:0.121299, cd_loss:0.024671, entropy_loss:0.000005, deform_loss:0.000153, lowrank_loss:0.001809
2022-04-11 07:22:31,845 : Batch 5800 Loss:0.124483, corr_loss:0.105013, cd_loss:0.017752, entropy_loss:0.000004, deform_loss:0.000135, lowrank_loss:0.001578
2022-04-11 07:22:38,269 : Batch 5810 Loss:0.136623, corr_loss:0.119391, cd_loss:0.015243, entropy_loss:0.000004, deform_loss:0.000133, lowrank_loss:0.001852
2022-04-11 07:22:43,064 : Batch 5820 Loss:0.159818, corr_loss:0.133095, cd_loss:0.024861, entropy_loss:0.000004, deform_loss:0.000162, lowrank_loss:0.001696
2022-04-11 07:22:49,981 : Batch 5830 Loss:0.159613, corr_loss:0.131518, cd_loss:0.026577, entropy_loss:0.000004, deform_loss:0.000117, lowrank_loss:0.001396
2022-04-11 07:22:55,398 : Batch 5840 Loss:0.127762, corr_loss:0.107937, cd_loss:0.018362, entropy_loss:0.000004, deform_loss:0.000152, lowrank_loss:0.001308
2022-04-11 07:23:00,962 : Batch 5850 Loss:0.127322, corr_loss:0.111806, cd_loss:0.014305, entropy_loss:0.000005, deform_loss:0.000112, lowrank_loss:0.001096
2022-04-11 07:23:08,073 : Batch 5860 Loss:0.129297, corr_loss:0.108440, cd_loss:0.019306, entropy_loss:0.000005, deform_loss:0.000124, lowrank_loss:0.001422
2022-04-11 07:23:15,776 : Batch 5870 Loss:0.167962, corr_loss:0.145520, cd_loss:0.021149, entropy_loss:0.000006, deform_loss:0.000134, lowrank_loss:0.001153
2022-04-11 07:23:21,810 : Batch 5880 Loss:0.197241, corr_loss:0.174802, cd_loss:0.021063, entropy_loss:0.000003, deform_loss:0.000139, lowrank_loss:0.001234
2022-04-11 07:23:27,260 : Batch 5890 Loss:0.216616, corr_loss:0.177835, cd_loss:0.036533, entropy_loss:0.000003, deform_loss:0.000244, lowrank_loss:0.002000
2022-04-11 07:23:32,242 : Batch 5900 Loss:0.163611, corr_loss:0.136137, cd_loss:0.026894, entropy_loss:0.000003, deform_loss:0.000138, lowrank_loss:0.000438
2022-04-11 07:23:39,344 : Batch 5910 Loss:0.268385, corr_loss:0.159586, cd_loss:0.011610, entropy_loss:0.000003, deform_loss:0.000119, lowrank_loss:0.097068
2022-04-11 07:23:44,910 : Batch 5920 Loss:0.350295, corr_loss:0.149164, cd_loss:0.016909, entropy_loss:0.000003, deform_loss:0.000231, lowrank_loss:0.183987
2022-04-11 07:23:50,413 : Batch 5930 Loss:0.920049, corr_loss:0.178489, cd_loss:0.022725, entropy_loss:0.000004, deform_loss:0.000168, lowrank_loss:0.718663
2022-04-11 07:23:56,567 : Batch 5940 Loss:0.179104, corr_loss:0.153191, cd_loss:0.023358, entropy_loss:0.000005, deform_loss:0.000153, lowrank_loss:0.002396
2022-04-11 07:24:02,931 : Batch 5950 Loss:0.152043, corr_loss:0.133964, cd_loss:0.015110, entropy_loss:0.000004, deform_loss:0.000144, lowrank_loss:0.002821
2022-04-11 07:24:10,418 : Batch 5960 Loss:0.175748, corr_loss:0.152540, cd_loss:0.020229, entropy_loss:0.000002, deform_loss:0.000125, lowrank_loss:0.002852
2022-04-11 07:24:15,222 : Batch 5970 Loss:0.557693, corr_loss:0.180952, cd_loss:0.024084, entropy_loss:0.000002, deform_loss:0.000139, lowrank_loss:0.352516
2022-04-11 07:24:20,400 : Batch 5980 Loss:0.164646, corr_loss:0.142908, cd_loss:0.019423, entropy_loss:0.000003, deform_loss:0.000136, lowrank_loss:0.002176
2022-04-11 07:24:27,004 : Batch 5990 Loss:0.167778, corr_loss:0.150446, cd_loss:0.014688, entropy_loss:0.000002, deform_loss:0.000142, lowrank_loss:0.002499
2022-04-11 07:24:32,110 : Batch 6000 Loss:0.151957, corr_loss:0.132350, cd_loss:0.017354, entropy_loss:0.000003, deform_loss:0.000154, lowrank_loss:0.002096
2022-04-11 07:24:38,853 : Batch 6010 Loss:0.161332, corr_loss:0.143597, cd_loss:0.015208, entropy_loss:0.000003, deform_loss:0.000141, lowrank_loss:0.002383
2022-04-11 07:24:44,674 : Batch 6020 Loss:0.146042, corr_loss:0.124359, cd_loss:0.019413, entropy_loss:0.000004, deform_loss:0.000153, lowrank_loss:0.002113
2022-04-11 07:24:50,090 : Batch 6030 Loss:0.167414, corr_loss:0.148456, cd_loss:0.016587, entropy_loss:0.000004, deform_loss:0.000136, lowrank_loss:0.002231
2022-04-11 07:24:56,130 : Batch 6040 Loss:0.245671, corr_loss:0.131757, cd_loss:0.022333, entropy_loss:0.000004, deform_loss:0.000146, lowrank_loss:0.091431
2022-04-11 07:25:02,632 : Batch 6050 Loss:0.158918, corr_loss:0.137141, cd_loss:0.019525, entropy_loss:0.000002, deform_loss:0.000154, lowrank_loss:0.002096
2022-04-11 07:25:08,729 : Batch 6060 Loss:0.144677, corr_loss:0.128974, cd_loss:0.012056, entropy_loss:0.000001, deform_loss:0.000139, lowrank_loss:0.003506
2022-04-11 07:25:14,971 : Batch 6070 Loss:0.163134, corr_loss:0.141428, cd_loss:0.018614, entropy_loss:0.000001, deform_loss:0.000136, lowrank_loss:0.002955
2022-04-11 07:25:20,348 : Batch 6080 Loss:0.178839, corr_loss:0.161275, cd_loss:0.014825, entropy_loss:0.000001, deform_loss:0.000150, lowrank_loss:0.002589
2022-04-11 07:25:28,763 : Batch 6090 Loss:0.177177, corr_loss:0.154025, cd_loss:0.020273, entropy_loss:0.000001, deform_loss:0.000151, lowrank_loss:0.002728
2022-04-11 07:25:34,005 : Batch 6100 Loss:0.205118, corr_loss:0.180742, cd_loss:0.021688, entropy_loss:0.000001, deform_loss:0.000155, lowrank_loss:0.002531
2022-04-11 07:25:39,137 : Batch 6110 Loss:0.140715, corr_loss:0.120587, cd_loss:0.017866, entropy_loss:0.000001, deform_loss:0.000132, lowrank_loss:0.002129
2022-04-11 07:25:44,117 : Batch 6120 Loss:0.121930, corr_loss:0.104700, cd_loss:0.014153, entropy_loss:0.000001, deform_loss:0.000138, lowrank_loss:0.002938
2022-04-11 07:25:51,788 : Batch 6130 Loss:0.159974, corr_loss:0.130274, cd_loss:0.022835, entropy_loss:0.000001, deform_loss:0.000146, lowrank_loss:0.006718
2022-04-11 07:25:57,249 : Batch 6140 Loss:0.192346, corr_loss:0.166914, cd_loss:0.022647, entropy_loss:0.000001, deform_loss:0.000156, lowrank_loss:0.002628
2022-04-11 07:26:02,346 : Batch 6150 Loss:0.189960, corr_loss:0.163396, cd_loss:0.024124, entropy_loss:0.000001, deform_loss:0.000140, lowrank_loss:0.002298
2022-04-11 07:26:07,871 : Batch 6160 Loss:0.145737, corr_loss:0.124859, cd_loss:0.017940, entropy_loss:0.000002, deform_loss:0.000154, lowrank_loss:0.002782
2022-04-11 07:26:15,325 : Batch 6170 Loss:0.130483, corr_loss:0.114265, cd_loss:0.012683, entropy_loss:0.000001, deform_loss:0.000130, lowrank_loss:0.003403
2022-04-11 07:26:21,753 : Batch 6180 Loss:0.147745, corr_loss:0.130941, cd_loss:0.014452, entropy_loss:0.000001, deform_loss:0.000145, lowrank_loss:0.002206
2022-04-11 07:26:27,523 : Batch 6190 Loss:0.179217, corr_loss:0.148284, cd_loss:0.027994, entropy_loss:0.000001, deform_loss:0.000142, lowrank_loss:0.002797
2022-04-11 07:26:32,387 : Batch 6200 Loss:0.177200, corr_loss:0.152850, cd_loss:0.022009, entropy_loss:0.000001, deform_loss:0.000142, lowrank_loss:0.002198
2022-04-11 07:26:40,009 : Batch 6210 Loss:0.172815, corr_loss:0.145709, cd_loss:0.024553, entropy_loss:0.000002, deform_loss:0.000142, lowrank_loss:0.002410
2022-04-11 07:26:44,898 : Batch 6220 Loss:0.169863, corr_loss:0.131859, cd_loss:0.017734, entropy_loss:0.000002, deform_loss:0.000137, lowrank_loss:0.020130
2022-04-11 07:26:51,031 : Batch 6230 Loss:0.192499, corr_loss:0.164317, cd_loss:0.024804, entropy_loss:0.000002, deform_loss:0.000145, lowrank_loss:0.003232
2022-04-11 07:26:56,936 : Batch 6240 Loss:0.145409, corr_loss:0.120767, cd_loss:0.021301, entropy_loss:0.000002, deform_loss:0.000129, lowrank_loss:0.003209
2022-04-11 07:27:04,429 : Batch 6250 Loss:0.161274, corr_loss:0.134427, cd_loss:0.023983, entropy_loss:0.000002, deform_loss:0.000143, lowrank_loss:0.002719
2022-04-11 07:27:09,629 : Batch 6260 Loss:0.197566, corr_loss:0.171911, cd_loss:0.022160, entropy_loss:0.000001, deform_loss:0.000148, lowrank_loss:0.003346
2022-04-11 07:27:14,962 : Batch 6270 Loss:0.449405, corr_loss:0.187930, cd_loss:0.023193, entropy_loss:0.000001, deform_loss:0.000150, lowrank_loss:0.238131
2022-04-11 07:27:21,387 : Batch 6280 Loss:0.286750, corr_loss:0.180243, cd_loss:0.018598, entropy_loss:0.000002, deform_loss:0.000141, lowrank_loss:0.087766
2022-04-11 07:27:27,619 : Batch 6290 Loss:0.224757, corr_loss:0.202550, cd_loss:0.018794, entropy_loss:0.000001, deform_loss:0.000152, lowrank_loss:0.003260
2022-04-11 07:27:33,079 : Batch 6300 Loss:0.144296, corr_loss:0.127126, cd_loss:0.013946, entropy_loss:0.000001, deform_loss:0.000147, lowrank_loss:0.003075
2022-04-11 07:27:39,827 : Batch 6310 Loss:0.174579, corr_loss:0.153421, cd_loss:0.017926, entropy_loss:0.000002, deform_loss:0.000130, lowrank_loss:0.003100
2022-04-11 07:27:45,572 : Batch 6320 Loss:0.240689, corr_loss:0.215314, cd_loss:0.020162, entropy_loss:0.000006, deform_loss:0.000137, lowrank_loss:0.005070
2022-04-11 07:27:51,225 : Batch 6330 Loss:0.206055, corr_loss:0.188849, cd_loss:0.013425, entropy_loss:0.000002, deform_loss:0.000144, lowrank_loss:0.003634
2022-04-11 07:27:56,587 : Batch 6340 Loss:0.191800, corr_loss:0.171729, cd_loss:0.016815, entropy_loss:0.000001, deform_loss:0.000131, lowrank_loss:0.003124
2022-04-11 07:28:04,025 : Batch 6350 Loss:0.241368, corr_loss:0.211277, cd_loss:0.026727, entropy_loss:0.000001, deform_loss:0.000145, lowrank_loss:0.003218
2022-04-11 07:28:09,405 : Batch 6360 Loss:0.217268, corr_loss:0.195405, cd_loss:0.017690, entropy_loss:0.000001, deform_loss:0.000133, lowrank_loss:0.004039
2022-04-11 07:28:14,819 : Batch 6370 Loss:0.182568, corr_loss:0.158691, cd_loss:0.020242, entropy_loss:0.000001, deform_loss:0.000124, lowrank_loss:0.003511
2022-04-11 07:28:20,880 : Batch 6380 Loss:0.185456, corr_loss:0.168582, cd_loss:0.012798, entropy_loss:0.000000, deform_loss:0.000130, lowrank_loss:0.003947\

opened by CodeLHY 0

Maybe something wrong in computing 3D IoU

def transform_coordinates_3d(coordinates, sRT):
    """
    Args:
        coordinates: [3, N]
        sRT: [4, 4]
    Returns:
        new_coordinates: [3, N]
    """
    assert coordinates.shape[0] == 3
    coordinates = np.vstack([coordinates, np.ones((1, coordinates.shape[1]), dtype=np.float32)])
    new_coordinates = sRT @ coordinates
    new_coordinates = new_coordinates[:3, :] / new_coordinates[3, :]
    return new_coordinates

def compute_3d_IoU(sRT_1, sRT_2, size_1, size_2, class_name_1, class_name_2, handle_visibility):
    """ Computes IoU overlaps between two 3D bboxes. """
    def asymmetric_3d_iou(sRT_1, sRT_2, size_1, size_2):
        noc_cube_1 = get_3d_bbox(size_1, 0)
        bbox_3d_1 = transform_coordinates_3d(noc_cube_1, sRT_1)
        noc_cube_2 = get_3d_bbox(size_2, 0)
        bbox_3d_2 = transform_coordinates_3d(noc_cube_2, sRT_2)

        bbox_1_max = np.amax(bbox_3d_1, axis=0)
        bbox_1_min = np.amin(bbox_3d_1, axis=0)
        bbox_2_max = np.amax(bbox_3d_2, axis=0)
        bbox_2_min = np.amin(bbox_3d_2, axis=0)

        overlap_min = np.maximum(bbox_1_min, bbox_2_min)
        overlap_max = np.minimum(bbox_1_max, bbox_2_max)

        # intersections and union
        if np.amin(overlap_max - overlap_min) < 0:
            intersections = 0
        else:
            intersections = np.prod(overlap_max - overlap_min)
        union = np.prod(bbox_1_max - bbox_1_min) + np.prod(bbox_2_max - bbox_2_min) - intersections
        overlaps = intersections / union
        return overlaps

When I debugged method asymmetric_3d_iou(), I found that bbox_3d_1 shape is (3, 8), as described in method transform_coordinates_3d(). Therefore, bbox_1_max shape is (8,), so are bbox_1_min, bbox_2_max, andbbox_2_min.

But I think their shapes should be (3,), which represented bounds of 3D boxes.

And I did a test using method asymmetric_3d_iou(). Results are as follows:

use bbox_1_max = np.amax(bbox_3d_1, axis=0) which output shape is (8,)

from lib.utils import compute_3d_IoU
import numpy as np
pose = np.eye(4, dtype=np.float32)
pose[:, 3] = np.array([1, 1, 1, 1])
print(compute_3d_IoU(pose, pose, [1, 1, 1], [0.5, 0.5, 0.5], 'laptop', 'laptop', 1))

## output : nan

use bbox_1_max = np.amax(bbox_3d_1, axis=1) which output shape is (3,)

from lib.utils import compute_3d_IoU
import numpy as np
pose = np.eye(4, dtype=np.float32)
pose[:, 3] = np.array([1, 1, 1, 1])
print(compute_3d_IoU(pose, pose, [1, 1, 1], [0.5, 0.5, 0.5], 'laptop', 'laptop', 1))

## output : 0.125

opened by yanjh97 1

How much GPU memory be need to run this project?

Great project, thanks you sharing your code. I try to run this project on my laptap, but out of memory. So can you tell me how much GPU memory be used when you eval and train, I will choose a proper device rerun this project.

opened by nicu233 1

Owner

Chen Kai

GitHub

Official PyTorch implementation of CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds

CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds Introduction This is the official PyTorch implementation of o

96 Dec 7, 2022

Super Pix Adv - Offical implemention of Robust Superpixel-Guided Attentional Adversarial Attack (CVPR2020)

Super_Pix_Adv Offical implemention of Robust Superpixel-Guided Attentional Adver

8 Oct 26, 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

169 Jan 7, 2023

The implemention of Video Depth Estimation by Fusing Flow-to-Depth Proposals

Flow-to-depth (FDNet) video-depth-estimation This is the implementation of paper Video Depth Estimation by Fusing Flow-to-Depth Proposals Jiaxin Xie,

32 Jun 14, 2022

(ImageNet pretrained models) The official pytorch implemention of the TPAMI paper "Res2Net: A New Multi-scale Backbone Architecture"

Res2Net The official pytorch implemention of the paper "Res2Net: A New Multi-scale Backbone Architecture" Our paper is accepted by IEEE Transactions o

928 Dec 29, 2022

The official pytorch implemention of the CVPR paper "Temporal Modulation Network for Controllable Space-Time Video Super-Resolution".

This is the official PyTorch implementation of TMNet in the CVPR 2021 paper "Temporal Modulation Network for Controllable Space-Time VideoSuper-Resolu

95 Oct 24, 2022

Official PyTorch implemention of our paper "Learning to Rectify for Robust Learning with Noisy Labels".

WarPI The official PyTorch implemention of our paper "Learning to Rectify for Robust Learning with Noisy Labels". Run python main.py --corruption_type

3 Sep 3, 2022

Repository for the paper "PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation", CVPR 2021.

PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation Code repository for the paper: PoseAug: A Differentiable Pose Augme

328 Dec 17, 2022

This repository contains codes of ICCV2021 paper: SO-Pose: Exploiting Self-Occlusion for Direct 6D Pose Estimation

SO-Pose This repository contains codes of ICCV2021 paper: SO-Pose: Exploiting Self-Occlusion for Direct 6D Pose Estimation This paper is basically an

52 Nov 25, 2022

[ICRA 2022] CaTGrasp: Learning Category-Level Task-Relevant Grasping in Clutter from Simulation

This is the official implementation of our paper: Bowen Wen, Wenzhao Lian, Kostas Bekris, and Stefan Schaal. "CaTGrasp: Learning Category-Level Task-R

199 Jan 4, 2023

Symmetry and Uncertainty-Aware Object SLAM for 6DoF Object Pose Estimation

SUO-SLAM This repository hosts the code for our CVPR 2022 paper "Symmetry and Uncertainty-Aware Object SLAM for 6DoF Object Pose Estimation". ArXiv li

Robot Perception & Navigation Group (RPNG)

97 Jan 3, 2023

[CVPR 2022] Pytorch implementation of "Templates for 3D Object Pose Estimation Revisited: Generalization to New objects and Robustness to Occlusions" paper

template-pose Pytorch implementation of "Templates for 3D Object Pose Estimation Revisited: Generalization to New objects and Robustness to Occlusions

92 Dec 28, 2022