Chengxin Liu

Hi, I have implemented a pytorch version of KL-Loss with MMDetection in my project [KL-Loss-pytorch](https://github.com/cxliu0/KL-Loss-pytorch). The reproduced results are as follows: | KL Loss | Var Vote | soft-NMS |...

Thanks for your interest. but we currently have no plan to release the implementation of Co-Teaching and SD-LocNet. Instead, the implementation details of Co-Teaching and SD-LocNet can be found in...

Feature extractors could be one of the reasons that limit the detection performance. In addition, I think the deployment of our approach on YOLOv5 is similar to OA-MIL FasterRCNN (please...

The implementations of OA-MIL are in these two files. No other modification is required.

Here is my workaround to run the model without connecting to huggingface: - Step 1: download necessary files listed in [huggingface-bert-base-uncased](https://huggingface.co/bert-base-uncased/tree/main), including ```config.json, flax_model.msgpack, pytorch_model.bin, tf_model.h5, tokenizer.json, tokenizer_config.json, vocab.txt``` -...

We did use ```coco2017_train.pkl``` to train clean-FasterRCNN and clean-RetinaNet, and this file works fine. We have not encountered this error before. Perhaps you were training the model in Windows system?...

We just used MMDetection to process the COCO dataset instead of separately installing pycocotools. In addition, the version of pytorch and python can be found [here](https://github.com/cxliu0/OA-MIL/tree/main#installation).