/AICity_Team24

Primary LanguagePythonMIT LicenseMIT

Effective Object Detection From Traffic Camera Videos

Our UIUC-IFP Team:

  • Honghui Shi
  • Zhichao Liu
  • Yuchen Fan
  • Xinchao Wang
  • Prof. Thomas Huang

Our Implementation for Nvidia AI City Challenge

Our implementation py-rfcn is adapted from py-R-FCN, with additions and modification to support our winning solution to the 1st IEEE Smart World Nvidia AI City Challenge. (For usage and installation of the original py-R-FCN, please refer to here.)

code usage

1. Prepare dataset

$ mkdir -p CODE_DIR/data/AICdevkit/results/AIC/Main
$ cd CODE_DIR/tools
$ python ./preprocess.py DATASET_DIR CODE_DIR/data

2. Build from source

$ sh compile.sh
$ export PYTHONPATH=$PYTHONPATH:CODE_DIR/caffe/python

3. Test on our pre-trained model

Download models from here.

Put ResNet-101-model.caffemodel in CODE_DIR/data/imagenet_models/

Put aic_trainval in CODE_DIR/output/rfcn_alt_opt_5step_ohem

$ bash test.sh 0

4. Train on aic dataset

We find it is better to train vehicle detector separately from traffic-signal detector.

if you want to train without traffic-signal

$ sh train.sh 0

if you want to train on traffic-signal

$ sh train.sh 1

5. Postprocess for submission

$ python postprocess.py CODE_DIR/data output_dir

Acknowledgement

This work is supported in part by IBM-ILLINOIS Center for Cognitive Computing Systems Research (C3SR) - a research collaboration as part of the IBM Cognitive Horizons Network.