search

lck1201 / win_det_heatmaps

Window Detection in Facade Using Heatmaps Fushion
MIT License
49 stars 18 forks source link

trainning error #10

Open zhu-xiao-di opened 2 years ago

zhu-xiao-di commented 2 years ago

Hello, thank you for your work.

When I train my own dataset, the following errors are reported in the verification phase. Can you help me?

in valid in eval Param Tag_threshold 12.0 Param detection_threshold 0.2 Traceback (most recent call last): File "/mnt/Code/win_det_heatmaps/common_pytorch/net_modules.py", line 152, in evalNet rectify = test_config.rectify, winScoreThres = test_config.windowT) File "/mnt/Code/win_det_heatmaps/common_pytorch/group/tag_group.py", line 266, in group_corners_on_tags grouped = parser.parse(np.float32(dets), np.float32(tags), idx, ratio, rectify) # shape=(num_of_windows, 4, 4) File "/mnt/Code/win_det_heatmaps/common_pytorch/group/tag_group.py", line 242, in parse re = self.calc(det, tag, idx) File "/mnt/Code/win_det_heatmaps/common_pytorch/group/tag_group.py", line 178, in calc val_k = [c_pts[:, 2, np.newaxis] for c_pts in coords_in_patch_with_score_id] File "/mnt/Code/win_det_heatmaps/common_pytorch/group/tag_group.py", line 178, in val_k = [c_pts[:, 2, np.newaxis] for c_pts in coords_in_patch_with_score_id] IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "train.py", line 158, in main() File "train.py", line 135, in main config.train.patch_width, config.train.patch_height, final_output_path) File "/mnt/Code/win_det_heatmaps/common_pytorch/net_modules.py", line 155, in evalNet assert 0, (n_s, e, os.path.basename(imdb_list[n_s]['image'])) AssertionError: (0, IndexError('too many indices for array: array is 1-dimensional, but 2 were indexed'), 'result_toushi_2282_6_4.jpg')

charvi-077 commented 9 months ago

Hi @zhu-xiao-di , were you able to resolve this issue ?

charvi-077 commented 9 months ago

Update : I figured out that this issue is same as related to issue #5 We need to have enough data to train again ( even in our case we are using both : our data and facade_dataset as provide in repository, to avoid this error ) , otherwise in evaluation it would not be able to detect the corner points for a given image and end up in this error.

Thanks.