search

zhiyuanyou / SAFECount

[WACV 2023] Few-shot Object Counting with Similarity-Aware Feature Enhancement
Other
124 stars 22 forks source link

Error in rop_roi_feat () , feat_box.shape[2] =0 #27

Closed makangzhe closed 1 year ago

makangzhe commented 1 year ago

Thanks for your amazing work ! I try to train this code im my own dataset , I generate train.json and test.json. and hvae checked bboxes and point is true , my orginal image size is (1920 ,1080） ， when i train this code, this error info image

i debug this code, find error happened in this function rop_roi_feat in models.utils def crop_roi_feat(feat, boxes, out_stride): """ feat: 1 x c x h x w boxes: m x 4, 4: [y_tl, x_tl, y_br, x_br] """ _, _, h, w = feat.shape boxes_scaled = boxes / out_stride boxes_scaled[:, :2] = torch.floor(boxes_scaled[:, :2]) # y_tl, x_tl: floor boxes_scaled[:, 2:] = torch.ceil(boxes_scaled[:, 2:]) # y_br, x_br: ceil boxes_scaled[:, :2] = torch.clamp_min(boxes_scaled[:, :2], 0) boxes_scaled[:, 2] = torch.clamp_max(boxes_scaled[:, 2], h) boxes_scaled[:, 3] = torch.clamp_max(boxes_scaled[:, 3], w) feat_boxes = [] for idx_box in range(0, boxes.shape[0]): y_tl, x_tl, y_br, x_br = boxes_scaled[idx_box] y_tl, x_tl, y_br, x_br = int(y_tl), int(x_tl), int(y_br), int(x_br) feat_box = feat[:, :, y_tl : (y_br + 1), x_tl : (x_br + 1)] if feat_box.shape[2] == 0: continue feat_boxes.append(feat_box) return feat_boxes in this function , (y_br + 1) > x_tl is the reason of feat_box.shape[2]=0 , how to solve this , please help me!

zhiyuanyou commented 1 year ago

Well, please give me the value of _boxes & boxesscaled of the specific failure case in this line (https://github.com/zhiyuanyou/SAFECount/blob/main/models/utils.py#L87).

makangzhe commented 1 year ago

Well, please give me the value of _boxes & boxesscaled of the specific failure case in this line (https://github.com/zhiyuanyou/SAFECount/blob/main/models/utils.py#L87).

boxes ：tensor([[825.3125, 233.9375, 856.1875, 262.4375], [312.3125, 165.0625, 377.6250, 203.0625], [837.1875, 233.9375, 868.0625, 245.8125]], device='cuda:0', dtype=torch.float64) boxes_scaled ： tensor([[206.3281, 58.4844, 214.0469, 65.6094], [ 78.0781, 41.2656, 94.4062, 50.7656], [209.2969, 58.4844, 217.0156, 61.4531]], device='cuda:0', dtype=torch.float64)

zhiyuanyou commented 1 year ago

For your data, I do not think the data have something wrong. How about h,w (https://github.com/zhiyuanyou/SAFECount/blob/main/models/utils.py#L86)?

makangzhe commented 1 year ago

For your data, I do not think the data have something wrong. How about h,w (https://github.com/zhiyuanyou/SAFECount/blob/main/models/utils.py#L86)?

feat.shape is (152,152) , https://github.com/zhiyuanyou/SAFECount/blob/de067f9f1ca2caea432dd4c2e6d9ec9b2a169ebf/models/utils.py#L93C1-L93C1 When the code runs to this line，boxes_scaled value is
tensor([[206., 58., 152., 66.], [ 78., 41., 95., 51.], [209., 58., 152., 62.]], device='cuda:0', dtype=torch.float64)

The value of Axis 2 is greater than the value of Axis 0, causing the Tensor to be sliced with a value 0 in axis 2

zhiyuanyou commented 1 year ago

This is not the error of code. It is because your annotation is out of image.

makangzhe commented 1 year ago

This is not the error of code. It is because your annotation is out of image.

Thanks for your reply, I solved this problem by adding these two lines of code in here https://github.com/zhiyuanyou/SAFECount/blob/de067f9f1ca2caea432dd4c2e6d9ec9b2a169ebf/models/utils.py#L92 , can I ask if these lines of code will affect the training results a lot?

boxes_scaled[:, 0] = torch.clamp_max(boxes_scaled[:, 0], h-1)
boxes_scaled[:, 1] = torch.clamp_max(boxes_scaled[:, 1], w-1)
zhiyuanyou commented 1 year ago

I think it is right for training.