box_ops matcher_ index out of bounds

[EvelyneCalista]

HI all, I work on custom dataset for object detction with bounding box only. I got below error " index 2 is out of bounds for dimension 1 with size 1" when doing matcher. does the width and height must be calculated in the dataset class? are there any people that got a similar problem with me? :(((((

def generalized_box_iou(boxes1, boxes2): assert (boxes1[:, 2:] >= boxes1[:, :2]).all() assert (boxes2[:, 2:] >= boxes2[:, :2]).all()

File "//detr-main/engine.py", line 34, in train_one_epoch loss_dict = criterion(outputs, targets) File "/home/anaconda3/envs/myenv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl return forward_call(*input, kwargs) File "detr-main/models/detr.py", line 225, in forward indices = self.matcher(outputs_without_aux, targets) File "/home/anaconda3/envs/myenv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl return forward_call(*input, *kwargs) File "/home/anaconda3/envs/myenv/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context return func(args, kwargs) File "/detr-main/models/matcher.py", line 74, in forward cost_giou = -generalized_box_iou(box_cxcywh_to_xyxy(out_bbox), box_cxcywh_to_xyxy(tgt_bbox)) File "/detr-main/util/box_ops.py", line 53, in generalized_box_iou iou, union = box_iou(boxes1, boxes2) File "/detr-main/util/box_ops.py", line 26, in box_iou area2 = box_area(boxes2) File "/home/anaconda3/envs/myenv/lib/python3.8/site-packages/torchvision/ops/boxes.py", line 229, in box_area return (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1]) IndexError: index 2 is out of bounds for dimension 1 with size 1

[truncs]

Check the polygons in the dataset. Might be helpful to print the id the sample in the dataset where this is erroring out.

[Sahar-DataScience]

I'm trying to use DETR for instance segmentation on a custom dataset in coco format and I faced the same issue

[reubenwenisch]

Facing the same issue when using custom coco dataset. Isn't coco format xywh but in the code thinks bbox is of the format cxcywh therefore they use box_cxcywh_to_xyxy. Is this the issue? Were any of you able to solve this?

[johannesSX]

But the conversion doesn't make any sense, does it? Version given in the code:

def box_xyxy_to_cxcywh(x):
    x0, y0, x1, y1 = x.unbind(-1)
    b = [(x0 + x1) / 2, (y0 + y1) / 2,
         (x1 - x0), (y1 - y0)]
    return torch.stack(b, dim=-1)

I think that's correct though:

def box_xyxy_to_cxcywh(x):
    x0, y0, x1, y1 = x.unbind(-1)
    b = [x0 + x1 / 2, y0 + y1 / 2,
         (x1 / 2), (y1 / 2)]
    return torch.stack(b, dim=-1)

I think the brackets were put in the wrong place when really the center of the box is supposed to be the reference coordinate. But it works, where is my mistake?

[aleenarayamajhi]

I have same problem in boxes1. I'm not getting negative values, but the values in boxes1 keep changing. Did you figure out the solution? @EvelyneCalista

     52     pdb.set_trace()
---> 53     assert (boxes1[:, 2:] >= boxes1[:, :2]).all()
     54     assert (boxes2[:, 2:] >= boxes2[:, :2]).all()
     55     iou, union = box_iou(boxes1, boxes2)

When I run: run F:/DETR/detr/main.py --dataset_file face --data_path F:/DETR/dataset/ --output_dir F:/DETR/output --resume F:/DETR/weights/detr-r50-e632da11.pth

My output looks like this:

Not using distributed mode
git:
  sha: 3c5317060f0c2ee3d1dccd3c4da23b3c2cb251bf, status: has uncommited changes, branch: master

Namespace(aux_loss=True, backbone='resnet50', batch_size=2, bbox_loss_coef=5, clip_max_norm=0.1, coco_panoptic_path=None, data_path='F:/DETR/dataset/', dataset_file='face', dec_layers=6, device='cuda', dice_loss_coef=1, dilation=False, dim_feedforward=2048, dist_url='env://', distributed=False, dropout=0.1, enc_layers=6, eos_coef=0.1, epochs=300, eval=False, frozen_weights=None, giou_loss_coef=2, hidden_dim=256, lr=0.0001, lr_backbone=1e-05, lr_drop=200, mask_loss_coef=1, masks=False, nheads=8, num_queries=30, num_workers=2, output_dir='F:/DETR/output', position_embedding='sine', pre_norm=False, remove_difficult=False, resume='F:/DETR/weights/detr-r50-e632da11.pth', seed=42, set_cost_bbox=5, set_cost_class=1, set_cost_giou=2, start_epoch=0, weight_decay=0.0001, world_size=1)
number of params: 41261318
IMAGE FOLDER:  F:\DETR\dataset\train
ANN_FILE:  F:\DETR\dataset\train.json
loading annotations into memory...
Done (t=0.32s)
creating index...
index created!
IMAGE FOLDER:  F:\DETR\dataset\val
ANN_FILE:  F:\DETR\dataset\val.json
loading annotations into memory...
Done (t=0.02s)
creating index...
index created!
Length of Dataset train:  500
Length of Dataset val:  100
Start training
Epoch: [0]  [  0/250]  eta: 0:11:30  lr: 0.000100  class_error: 2.00  loss: 36.0587 (36.0587)  loss_ce: 0.3236 (0.3236)  loss_bbox: 4.4387 (4.4387)  loss_giou: 1.5512 (1.5512)  loss_ce_0: 0.2778 (0.2778)  loss_bbox_0: 3.5100 (3.5100)  loss_giou_0: 1.8308 (1.8308)  loss_ce_1: 0.2402 (0.2402)  loss_bbox_1: 3.9349 (3.9349)  loss_giou_1: 1.7132 (1.7132)  loss_ce_2: 0.2037 (0.2037)  loss_bbox_2: 4.0667 (4.0667)  loss_giou_2: 1.6071 (1.6071)  loss_ce_3: 0.2174 (0.2174)  loss_bbox_3: 4.4361 (4.4361)  loss_giou_3: 1.5737 (1.5737)  loss_ce_4: 0.2624 (0.2624)  loss_bbox_4: 4.3035 (4.3035)  loss_giou_4: 1.5675 (1.5675)  loss_ce_unscaled: 0.3236 (0.3236)  class_error_unscaled: 2.0000 (2.0000)  loss_bbox_unscaled: 0.8877 (0.8877)  loss_giou_unscaled: 0.7756 (0.7756)  cardinality_error_unscaled: 33.5000 (33.5000)  loss_ce_0_unscaled: 0.2778 (0.2778)  loss_bbox_0_unscaled: 0.7020 (0.7020)  loss_giou_0_unscaled: 0.9154 (0.9154)  cardinality_error_0_unscaled: 34.0000 (34.0000)  loss_ce_1_unscaled: 0.2402 (0.2402)  loss_bbox_1_unscaled: 0.7870 (0.7870)  loss_giou_1_unscaled: 0.8566 (0.8566)  cardinality_error_1_unscaled: 34.0000 (34.0000)  loss_ce_2_unscaled: 0.2037 (0.2037)  loss_bbox_2_unscaled: 0.8133 (0.8133)  loss_giou_2_unscaled: 0.8036 (0.8036)  cardinality_error_2_unscaled: 34.0000 (34.0000)  loss_ce_3_unscaled: 0.2174 (0.2174)  loss_bbox_3_unscaled: 0.8872 (0.8872)  loss_giou_3_unscaled: 0.7869 (0.7869)  cardinality_error_3_unscaled: 34.0000 (34.0000)  loss_ce_4_unscaled: 0.2624 (0.2624)  loss_bbox_4_unscaled: 0.8607 (0.8607)  loss_giou_4_unscaled: 0.7838 (0.7838)  cardinality_error_4_unscaled: 34.0000 (34.0000)  time: 2.7605  data: 2.1511  max mem: 3155
Traceback (most recent call last):

  File "F:\DETR\detr\main.py", line 259, in <module>
    main(args)

  File "F:\DETR\detr\main.py", line 209, in main
    args.clip_max_norm)

  File "F:\DETR\detr\engine.py", line 33, in train_one_epoch
    loss_dict = criterion(outputs, targets)

  File "C:\ProgramData\Anaconda3\envs\demo\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)

  File "F:\DETR\detr\models\detr.py", line 224, in forward
    indices = self.matcher(outputs_without_aux, targets)

  File "C:\ProgramData\Anaconda3\envs\demo\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)

  File "C:\ProgramData\Anaconda3\envs\demo\lib\site-packages\torch\autograd\grad_mode.py", line 15, in decorate_context
    return func(*args, **kwargs)

  File "F:\DETR\detr\models\matcher.py", line 74, in forward
    cost_giou = -generalized_box_iou(box_cxcywh_to_xyxy(out_bbox), box_cxcywh_to_xyxy(tgt_bbox))

  File "F:\DETR\detr\util\box_ops.py", line 51, in generalized_box_iou
    assert (boxes1[:, 2:] >= boxes1[:, :2]).all()

AssertionError

Please help!!