ke1337
commented
6 years ago Can you print out shape of bboxes and label_idxs? Seems like a bug in data loading. Please send a PR if you have a fix for this.
Closed JONYJONS closed 6 years ago
ke1337
commented
6 years ago Can you print out shape of bboxes and label_idxs? Seems like a bug in data loading. Please send a PR if you have a fix for this.
JONYJONS
commented
6 years ago this is example of bboxes:
294 705 394 805
This is for label:
car
Both are TSV files, which bring me to another question is this the only file type that is allowed?
The program gives error only, when there is a single object in the bboxes and label.
JONYJONS
commented
6 years ago Hi, sorry for the late replay.
The problem caused because I manually changed the data in the TSV files in the grocery model in faster rcnn, to use with my data. These files didn't have an automatic new line operator at the end of a line. This caused the errors when
having only one object. I used other TSV files and the problem was solved.
From: KeDengMS notifications@github.com Sent: Thursday, May 31, 2018 11:32 PM To: Microsoft/CNTK Cc: JONYJONS; Author Subject: Re: [Microsoft/CNTK] OBJECT DETECTION (#3227)
Can you print out shape of bboxes and label_idxs? Seems like a bug in data loading. Please send a PR if you have a fix for this.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/Microsoft/CNTK/issues/3227#issuecomment-393714440, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AkskezytCAQQnTk7w97_sVXkQ2TETIrVks5t4H2RgaJpZM4UQIXY.
spandantiwari
commented
6 years ago OK, thanks for letting us know.
There is an error in the training stage, when the input image data (bounding boxes or labes, files)appears a single object, in CNTK version 2.2.
The error doesn't appear when there is more than one object in the input image data.
The error message is, C:\Users\USER\AppData\Local\Programs\Python\Python35\Lib\site-packages\CNTK-Samples-2-2\Examples\Image\Detection\utils\annotations>python annotations_helper.py Traceback (most recent call last): File "annotations_helper.py", line 137, in
create_map_files(data_set_path, class_dict, training_set=True)
File "annotations_helper.py", line 75, in create_map_files
gt_annotations = _load_annotation(abs_img_path, class_dict)
File "annotations_helper.py", line 59, in _load_annotation
annotations = np.hstack((bboxes, label_idxs))
File "C:\Users\USER\AppData\Local\Programs\Python\Python35\lib\site-packages\numpy\core\shape_base.py", line 288, in hstack
return _nx.concatenate(arrs, 1)
ValueError: all the input array dimensions except for the concatenation axis must match exactly