Closed SherlockHolmes221 closed 3 years ago
More question: Could you provide the training log files? How many GPUs and the batch sizes did you use to get the final results in paper? Thanks
Please check the closed issue to see if it works for you: issues_7. Sorry we currently don't have the training logs. We train with 32 GPUs and 2 sample for each GPU. You can decrease the GPU amounts according to your devices, and then modify the learning rate correspondingly, e.g.: 32 GPU 2 img/per GPU(batchsize 64) & lr -> 8 GPU 2 img/per GPU(batchsize 16) & 0.25lr.
Thanks
running the test code on HICO-DET python3 tools/eval.py --cfg configs/hico.yaml