Hi, I have followed all your steps on demo.ipynb, but when I run the final command, it show this kind of error. How do I solve it. And is there new version for CUDA 10.1 10.2 or 9.1 to run this code.

The nuclei dateset with annotation file can be found here: https://github.com/hrlblab/CircleNet/blob/master/docs/DATA.md

1 在COCO格式的注释里，annotations中的segmentation字段多出了如下图所示的内容，其中width和height在image字段中已经出现了，如果直接用circlenet直接训练的话，这个必须加吗？

2 extreme_points代表什么意思？ 3 MoNuSeg_extreme_train2021.json和MoNuSeg_train2021.json有什么区别？

Hello, thank you for your code sharing. I encountered the following problems when training my dataset: first, the training loss seems to be the same as that of centernet, with no circle loss. Second, Why is cIOU not reflected in loss. Hope to get your answer, thank you.

Hi, I realized that my validation loss is increasing over time after a certain epoch while the training loss is converging. It seems to me that the network is overfitting. I have tried disabling random cropping, and introduced augmentations to mitigate this issue, however the problem seems to remain. I'm fairly new to neural network training, any suggestion is much appreciated!

Here is my opt file:

==> cudnn version: 7104
==> Cmd:
['main.py', 'circledet', '--exp_id', 'Apple5', '--arch', 'hourglass', '--batch_size', '4', '--master_batch', '4', '--lr', '2.5e-4', '--dataset', 'apple2', '--load_model', '/home/lmga-titanx/CircleNet/src/lib/models/ctdet_coco_hg.pth', '--num_epochs', '1000', '--input_w', '128', '--input_h', '128', '--K', '1', '--not_rand_crop', '--rotate', '1']
==> Opt:
  K: 1
  aggr_weight: 0.0
  agnostic_ex: False
  arch: hourglass
  aug_ddd: 0.5
  aug_rot: 0
  batch_size: 4
  cat_spec_wh: False
  center_thresh: 0.1
  chunk_sizes: [4]
  data_dir: /home/lmga-titanx/CircleNet/src/lib/../../data
  dataset: apple2
  debug: 0
  debug_dir: /home/lmga-titanx/CircleNet/src/lib/../../exp/circledet/Apple5/debug
  debugger_theme: white
  demo: 
  demo_dir: 
  dense_hp: False
  dense_wh: False
  dep_weight: 1
  dim_weight: 1
  down_ratio: 4
  eval_oracle_dep: False
  eval_oracle_hm: False
  eval_oracle_hmhp: False
  eval_oracle_hp_offset: False
  eval_oracle_kps: False
  eval_oracle_offset: False
  eval_oracle_wh: False
  exp_dir: /home/lmga-titanx/CircleNet/src/lib/../../exp/circledet
  exp_id: Apple5
  ez_guassian_radius: False
  filter_boarder: False
  fix_res: True
  flip: 0.5
  flip_test: False
  gpus: [0]
  gpus_str: 0
  head_conv: 64
  heads: {'hm': 1, 'cl': 1, 'reg': 2}
  hide_data_time: False
  hm_hp: True
  hm_hp_weight: 1
  hm_weight: 1
  hp_weight: 1
  input_h: 128
  input_res: 128
  input_w: 128
  keep_res: False
  kitti_split: 3dop
  load_model: /home/lmga-titanx/CircleNet/src/lib/models/ctdet_coco_hg.pth
  lr: 0.00025
  lr_step: [90, 120]
  lv: 2
  mask_focal_loss: False
  master_batch_size: 4
  mean: [[[0.40789655 0.44719303 0.47026116]]]
  metric: loss
  mse_loss: False
  nms: False
  no_color_aug: False
  norm_wh: False
  not_cuda_benchmark: False
  not_hm_hp: False
  not_prefetch_test: False
  not_rand_crop: True
  not_reg_bbox: False
  not_reg_hp_offset: False
  not_reg_offset: False
  num_classes: 1
  num_epochs: 1000
  num_iters: -1
  num_stacks: 2
  num_workers: 4
  off_weight: 1
  ontestdata: False
  output_h: 32
  output_res: 32
  output_w: 32
  pad: 127
  peak_thresh: 0.2
  print_iter: 0
  rect_mask: False
  reg_bbox: True
  reg_hp_offset: True
  reg_loss: l1
  reg_offset: True
  resume: False
  root_dir: /home/lmga-titanx/CircleNet/src/lib/../..
  rot_weight: 1
  rotate: 1.0
  rotate_reproduce: 0
  save_all: False
  save_dir: /home/lmga-titanx/CircleNet/src/lib/../../exp/circledet/Apple5
  scale: 0.4
  scores_thresh: 0.1
  seed: 317
  shift: 0.1
  std: [[[0.2886383  0.27408165 0.27809834]]]
  task: circledet
  test: False
  test_scales: [1.0]
  test_val_name: 
  trainval: False
  val_intervals: 1
  vis_thresh: 0.2
  wh_weight: 0.1```

If the dataset is inconvenient to publish, I hope to give pointers on the annotation of the dataset. It is best to provide an annotation file, thank you！！！

for example

Our dataset no circle_center and circle_radius fields in json file {"id": 1326, "image_id": 12, "category_id": 1, "iscrowd": 0, "area": 693, "bbox": [303.0, 270.0, 38.0, 30.0], "segmentation": [[337.0, 299.5, 315.0, 297.5, 305.0, 293.5, 302.5, 284.0, 307.5, 271.0, 314.0, 270.5, 324.0, 282.5, 331.0, 284.5, 333.0, 279.5, 340.5, 293.0, 337.0, 299.5]], "width": 500, "height": 500}

The monuseg2021 dataset you provided have circle_center and circle_radius fields {"id": 65786, "image_id": 309, "category_id": 1, "iscrowd": 0, "area": 191, "bbox": [393.0, 305.0, 19.0, 14.0], "segmentation": [[410.0, 318.5, 395.0, 314.5, 392.5, 309.0, 398.0, 304.5, 405.0, 305.5, 411.5, 312.0, 410.0, 318.5]], "width": 512, "height": 512, "circle_center": [401.583984375, 313.134765625], "circle_radius": 9.980734386753602, "extreme_points": [[398, 304], [392, 309], [410, 318], [411, 312]]}

Hi, I was trying to train CircleNet on my custom dataset with the image resolution of 128. I managed to train the NN with this following command: python main.py circledet --exp_id Apple2 --arch hourglass --batch_size 4 --master_batch 4 --lr 2.5e-4 --dataset apple --load_model /home/lmga-titanx/CircleNet/src/lib/models/ctdet_coco_hg.pth --input_res 128 --input_h 128 --input_w 128 --num_epochs 200

However, when I tried to run detection with my trained model using the following script:

TASK = 'circledet'
opt = opts().init('{} --dataset {} --load_model {} --arch {} --input_w {} --input_h {} --K {}'.format(TASK, 'coco', MODEL_PATH, 'hourglass', 128,128, 1).split(' '))

detector = detector_factory[opt.task](opt)
ret = detector.run(img)['results']

It returns an output that seemingly corresponds to the resolution of 512:

{1: array([[356.9697   , 344.58255  , 263.82367  ,   0.7991894,   0.       ]],
      dtype=float32)}

Can you please let me know what I may have done incorrectly? Thank you!

Now, I have my own COCO format dataset. I would like to train my own dataset. I intend to know the specific of command.

https://github.com/hrlblab/CircleNet/blob/master/experiments/circledet_monuseg_dla.sh In circledet_monuseg_dla.sh, what's the meaning of exp_id? I have searched the key word "CircleNet_DLA_Reproduce" in the whole project, but I can't find the related code.

I am running into an error when I run this command in the circle_net_demo.ipynb file

!python run_detection_for_scn.py circledet --arch dla_34 --demo "/content/drive/My Drive/CircleNet/example_case/Case 03-1.scn" --load_model /content/drive/My\ Drive/CircleNet/trained_model/model_10.pth --filter_boarder --demo_dir "/content/drive/My Drive/CircleNet/demo_dir/test_demo"

Here is the traceback

Traceback (most recent call last):
  File "run_detection_for_scn.py", line 470, in <module>
    demo(opt)
  File "run_detection_for_scn.py", line 461, in demo
    run_one_scn(demo_scn, demo_dir, opt)
  File "run_detection_for_scn.py", line 241, in run_one_scn
    detector = Detector(opt)
  File "/content/drive/MyDrive/CircleNet/src/lib/detectors/circledet.py", line 23, in __init__
    super(CircledetDetector, self).__init__(opt)
  File "/content/drive/MyDrive/CircleNet/src/lib/detectors/base_detector.py", line 25, in __init__
    self.model = load_model(self.model, opt.load_model)
  File "/content/drive/MyDrive/CircleNet/src/lib/models/model.py", line 34, in load_model
    checkpoint = torch.load(model_path, map_location=lambda storage, loc: storage)
  File "/usr/local/lib/python3.7/site-packages/torch/serialization.py", line 356, in load
    f = open(f, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: '/content/drive/My Drive/CircleNet/trained_model/model_10.pth'