Domain Contrast

Code of Domain Contrast for Domain Adaptive Object Detection, accepted in IEEE Transactions on Circuits and Systems for Video Technology(TCSVT)，2021.

The code is built based on the faster-rcnn. Please follow original project respository to set up the environment.

Data Preparation

• PASCAL_VOC 07+12: Please refer py-faster-rcnn for constructing PASCAL VOC Datasets.
• Clipart, Comic, WaterColor: Please refer Cross Domain Detection .
• SIM10k: Please refer website SIM10k
• Cityscape:Please refer website Cityscape, see dataset preparation code in DA-Faster RCNN
• Transferred Datasets: We use CycleGAN to generate transferred images.We trained CycleGAN with a learning rate of 2e-4 for the first ten epochs and a linear decaying rate to zero over the next ten epochs.

All codes are written to fit for the Data format of Pascal VOC. After downloading/generating the data, creat softlinks in the folder data/.

Pretrained Model

In our experiments, we used two pre-trained models on ImageNet, i.e., VGG16 and ResNet101. Please download these two models from:

• VGG16: Dropbox VT Server

• ResNet101: Dropbox VT Server

Download them and put them into the data/pretrained_model/.

Train and Test

All specific hyperparameters are in the shell scripts. Run with the following commands and you will get the results. Pascal2Clipart:

bash pascal2clipart.sh

Pascal2Comic:

bash pascal2comic_vgg16.sh

bash pascal2comic_resnet101.sh

Pascal2Watercolor:

bash pascal2watercolor.sh

SIM10K2Cityscape:

bash sim10k2city.sh

Results

Task	Backbone	mAP
Pascal2Clipart	Resnet101	43.2
Pascal2Comic	VGG16	36.9
Pascal2Watercolor	Resnet101	53.7
SIM10K2Cityscape	VGG16	41.6