jimmyli/faster-rcnn-gpu repository overview

This is a docker image for GPU-enabled Faster RCNN (https://github.com/rbgirshick/py-faster-rcnn⁠). It extends from nvidia/cuda:7.5-cudnn4-devel-ubuntu14.04. Run using nvidia-docker (https://github.com/NVIDIA/nvidia-docker⁠).

The demo has been modified to run without display. Both the VOC and Imagenet models are included in the image. To run it, do:

sudo nvidia-docker run --rm -it jimmyli/faster-rcnn-gpu /bin/bash cd /workspace/py-faster-rcnn/tools/ python demo.py (uses model refined on Pascal) python demo_coco.py (uses model refined on Coco)

The results can be found in /workspace/py-faster-rcnn/tools/demo_results

This image has been tested on host with Ubuntu 14.04 and GTX 770.

---

⁠Notes

Using GTX 770, the faster rcnn source code outputs the following error by default:

Check failed: error == cudaSuccess (8 vs. 0) invalid device function

This problem was reported here (https://github.com/rbgirshick/py-faster-rcnn/issues/2⁠). To resolve it, I used alantrrs's recommendation, and made sure to use -arch=sm_30 instead of -arch=sm_35 in /workspace/py-faster-rcnn/lib/setup.py

You will see the following error when importing caffe in python:

libdc1394 error: Failed to initialize libdc1394

It doesn't affect the demo code, but see the following pages for more info on this error:

https://groups.google.com/forum/#!topic/digits-users/uvQpHooD6WY⁠ http://stackoverflow.com/questions/12689304/ctypes-error-libdc1394-error-failed-to-initialize-libdc1394/26028597#26028597⁠ http://stackoverflow.com/questions/31768441/how-to-persist-ln-in-docker-with-ubuntu⁠

---

⁠Model refined on Coco

To get the demo using the model refined on Coco, I obtained the model from https://github.com/rbgirshick/py-faster-rcnn/blob/master/models/README.md⁠ and followed suggestions by enderhsu (https://github.com/rbgirshick/py-faster-rcnn/issues/381⁠) and liuchang8am (https://github.com/rbgirshick/py-faster-rcnn/issues/101⁠)