search

kienduynguyen / BoxeR

Code release for "BoxeR: Box-Attention for 2D and 3D Transformers"
MIT License
139 stars 22 forks source link

How to use the retrained model to evaluate on my own test dataset #12

Closed imhumanbean closed 2 years ago

imhumanbean commented 2 years ago

when i evaluate on my own test dataset, the evaluation looks not work. I just changed the data path on config file. 2022-07-26T10:17:57 INFO: Total Parameters: 39696075. Trained Parameters: 39696075 2022-07-26T10:17:57 INFO: Starting inference on test set 2022-07-26T10:17:57 INFO: Evaluation time. Running on full test set... 2022-07-26T10:19:01 INFO: progress on test: 0/0, : --------------------, update: 0, epoch: 0, max_update: 0, num_image: 873 2022-07-26T10:19:02 INFO: The inference finished!

kienduynguyen commented 2 years ago

If you wanna see the COCO metrics evaluation, you should set it to val mode. The test mode generates a json file that can be used to submit to the COCO server.

imhumanbean commented 2 years ago

thanks a lot. I calculated the metrics I needed from the output JSON file.