import error in augmentors.py

[anbilly19]

Traceback (most recent call last): File "train_pose1.py", line 17, in <module> from training.dataset import get_dataflow, batch_dataflow File "../training/dataset.py", line 11, in <module> from training.augmentors import ScaleAug, RotateAug, CropAug, FlipAug, \ File "../training/augmentors.py", line 6, in <module> from tensorpack.dataflow.imgaug.geometry import WarpAffineTransform,TransformAugmentorBase ImportError: cannot import name TransformAugmentorBase

[michalfaber]

Yes, the code is not compatible with the latest version of tensorpack. Try to use tensorpack==0.9.0.1 Thanks for the info.

[anbilly19]

Thanks for the exact stable version i was unable to find the correct one. But now i'm seeing a ZeroDivision error in ScaleAug line 226. while the code proceeds to the first epoch it shows the same error and gets stuck midway. I'm not sure of the cause tho. Should i start a new issue for this lemme know

[michalfaber]

Are you using the annotation file person_keypoints_train2017.json ? If not, there should be bbox element in your json file, something like "bbox": [x, y, w, h]

[anbilly19]

Yes I'm using the same one. Download from the coco website.

On Tue, 10 Sep 2019, 4:25 pm Michal Faber, [email protected] wrote:

Are you using the annotation file person_keypoints_train2017.json ? If not, there should be bbox element in your json file, something like "bbox": [x, y, w, h]

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/michalfaber/keras_Realtime_Multi-Person_Pose_Estimation/issues/121?email_source=notifications&email_token=AG7E4NHXRTLEZGIJ4ESHIBTQI54LTA5CNFSM4ISN6IHKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6KV3ZQ#issuecomment-529882598, or mute the thread https://github.com/notifications/unsubscribe-auth/AG7E4NERG7VXXG3F7C74NJ3QI54LTANCNFSM4ISN6IHA .

[Deeptanshu-sankhwar]

@anbilly19 try unzipping wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip in folder annotations. Worked for me !!