Unsupervised domain adaptation architercture for pedestrian detection. See the paper for details or the ppt for a demonstration.
Some new layers used for this project are implemented in my forked ApolloCaffe. Please firstly pull and compile the ApolloCaffe. If you have problems installing ApolloCaffe, you can refer to http://apollocaffe.com.
$ git clone https://github.com/LihangLiu93/apollocaffe.git
With ApolloCaffe installed, you can run ReInspect with:
$ git clone https://github.com/LihangLiu93/ReInspect.git
$ cd reinspect
The data consists of:
-
The source domain data from Russell's project, which this project is built on. The data can be found here.
-
The target domain data collected for domain adaptation, please put the data from the following links into the corresponding directories.
dir:./multi_scene_data/annnotation/second_carteen/ link: https://pan.baidu.com/s/1c16O1Hm password: d9kc
dir:./multi_scene_data/pre_data/images_640_480/second_carteens[01-03]/ link: https://pan.baidu.com/s/1dFqhGWh password: rs9b
$ python train_boost_ip_split.py --gpu=0 --config=config_boost_ip_split.json --weights=./data/brainwash_800000.h5
If you have further questions regarding the project, please email backchord at gmail dot com.