faster-rcnn.pytorch
faster-rcnn.pytorch copied to clipboard
A faster pytorch implementation of faster r-cnn
Hi everyone. I'm pretty sure this repo is not maintained, you clearly stated in the readme :) But it looks like some people still try to use it, including me....
I find that pytorch pretrain model is low almost 3 precent than caffe model. Is it true?
when use pretrained model with Rsenet152,it occurs:  I guess it should be 12 anchors in rpn, but when i check cfg i found that there are 9 anchors ...
Hi everyone! I don't understand the instruction "choose the right -arch in the make.sh file", and I'm also confused by the links that the author shared ([this](https://developer.nvidia.com/cuda-gpus) and [this](http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/) website)....
did it support higher pytorch version like pytorch 1.7.1??
I have been train the model my custom dataset. I have been trained up to 50 epoch. The training loss does not decrease around 1.6. The training is going well...
I have tried the default pascal_classes, and MS COCO classes by changing some settings and add the 80 classes of MS COCO to demo.py. How should I do to detect...
no cython_fbbox file and i do not know how to use .pyx type file
https://github.com/jwyang/faster-rcnn.pytorch/blob/f9d984d27b48a067b29792932bcb5321a39c1f09/lib/datasets/imdb.py#L14 为什么在lib/utils/文件夹下没有对应的cython_bbox文件呢