Aditya Arun

@ZhouYanzhao @FzuLJ can you please share the code snippet (or detailed instructions) to convert the *.mat files (from MCG/COB) to JSON format?

Updated the code base for torchvision 0.3.0. Now the compilation of various ROI layers is not required.

@philokey I haven't tested it completely yet. With res101-C4 I was getting 0.751 mAP which looked reasonable. I have to run proper benchmarks though for various settings. Can you provide...

Hi @Hsintao, data for VOC needs to be in COCO's JSON format. Read more about preparing data [here](https://github.com/adityaarun1/Detectron.pytorch#data-preparation). P.S.: To answer the previous question, the setting for which I got...

@szrlee I used 60,000th iteration as my final model.

I believe, an Imagenet pretrained res101/vgg16 model is essential for the network to converge successfully. If you want to train your network from scratch (i.e. using a randomly initialized res101...

`RuntimeError: Not compiled with GPU support (nms at /home/janvin/faster-rcnn.pytorch/lib/model/csrc/nms.h:22)` NMS seems to be compiled without GPU. Try removing the `build` directory and recompiling. Ensure you have CUDA/CuDNN in your path,...

See #126

I have not implemented mask evaluation for VOC dataset. I checked and it is not there in the official [Detectron repo](https://github.com/facebookresearch/Detectron) as well. However, it should indeed possible to compute...

Thanks. The `master` branch already has this fixed [[refer here](https://github.com/adityaarun1/Detectron.pytorch/blob/master/lib/modeling/model_builder.py#L279-L280)]. The `pytorch-1.0` branch is obsolete.