Motion Segmentation is the task of identifying the independently moving objects (pixels) in the video and separating them from the background motion.
Thanks for helping from Chenqiu, and Dr.Basu.
Our work include use the state of the art : Pixel distribution model (random feature selection) to preprocess the extracted optical flow images. The optical flow is extracted use the state of the art model, RAFT provided by Teed and Deng. Then we will serve the extracted pixel distributions as inputs to our motion segmentation net. Finally, we will generate our own segementation binary masks against groud truths.
This is only a portion of the code , that do inculde the latest version. We will upload the managed/imporved version shortly.
Tested dataset: http://www.cvlibs.net/datasets/kitti/
To train/test the model, you need to download the dataset from:
We also test our model on DAVIS 2016:
To train the model,
python3 train.py --data_root=directory to the KITTI MoSeg --checkpoint=./ckptTo test the model from a saved checkpoint: Example
python3 test.py --data_root=directory --model_path=./checkpoint_0423/ckpt_22.pth --test_image_dir=the directory where you want to save tested image result --gt_dir=the directory to save the ground truththe pre-trained model is saved in /checkpoint_0423/ckpt_22.pth