sliu

We followed the ensemble techniques following several competition entries in ImageNet and COCO object detection track. For more details, please refer to the "Deep Residual Learning for Image Recognition ".

This codebase is heavily based on the Detectron.pytorch, which supports the multi-scale testing. So you can also conduct multi-scale testing with this codebase by adding more scales to the config...

Please use the flag --multi-gpu-testing at inference time.

I used 8 * P40 for this experiment with 2 images on one P40. P40 is with 24GB memory compared with the 16GB memory of P100. Maybe you can use...

Hi, Thanks a lot for your interest. Actually, this is a long story. To be honest, I firstly tried the PANet on another codebase which is not released currently. With...

Hi, I have uploaded the model trained with 2fc. The performance should be 39.6 box AP. Please try with the new model and corresponding config file. Thanks!

@JiamingSuen Thanks a lot for your interest! Sorry that I really don't have much time on this currently. Maybe you can try this by yourself since the pre-trained model on...

@JrPeng In our origin implementation, we used Sync BN. In this version, we us GN instead. Using this kind of normalization can help the network converge better and achieve better...

@JiamingSuen Sorry for the late reply. We have listed several hyper-parameters in our paper, which are borrowed from Mask R-CNN. In other words, we strictly followed the training parameters used...

Hi guys, thanks a lot for your interests! This codebase is heavily based on Detectron.pytorch by Roy. In this codebase and released configs, I used multi-scale training larger testing scale...