search

MedChaabane / DEFT

Joint detection and tracking model named DEFT, or ``Detection Embeddings for Tracking." Our approach relies on an appearance-based object matching network jointly-learned with an underlying object detection network. An LSTM is also added to capture motion constraints.
MIT License
268 stars 43 forks source link

Training is stuck at the beginning on custom dataset #17

Open rvrsprdx opened 3 years ago

rvrsprdx commented 3 years ago

Hi and thanks for your work! I'm trying to train DEFT on a dataset which I have used to train CenterTrack successfully before. When running train.py the output seems promising: Creating model... Setting up train data... ==> initializing train data from ../data/s/tracking_train.json, images from ../data/data_tracking_image/training/ ... loading annotations into memory... Done (t=0.18s) creating index... index created! Creating video index! Loaded Custom dataset 202 samples Starting training... yesyes tracking/nameOfMyDataset Here the process gets stuck and nothing happens. I can see that about 800mb has been occupied by my gpu.

Do you have any idea what might be causing this issue? Thanks!

anidh commented 2 years ago

Hi @rvrsprdx were you able to solve this issue?

rvrsprdx commented 2 years ago

Yes, Iirc I had to change the import order of pytorch and opencv. Surprisingly, that fixed the issue.

anidh commented 2 years ago

Thanks a lot @rvrsprdx for the quick response. Were you able to detect anything with the custom dataset training? When I'm using --debug option it's Time throwing me error.

Shelton-Zhou commented 1 year ago

Same question. Could you please tell me where the modifications should be made? I have changed the import order of pytorch and opencv in the test.py and other files, but it didn't work.