- Sanjeev Kumar Singh
- Ankur Kunder
Results on vot2013/bicycle:
Models | FPS |
---|---|
ADNet | 2.90 |
ADNet-Fast | 15.00 |
Ours | 9.65 |
Python version: This code is in Python3.6
Package Requirements: tensorflow==1.14.0, OpenCV==3.4.2, pyyaml, hdf5storage, h5py
- The plots shown are the obsrvations collected on single training video
Implementation of 'Action-Decision Networks for Visual Tracking with Deep Reinforcement Learning(CVPR 2017)'
GIF | Description |
---|---|
vot2013 Dataset bicycle |
- Green : Ground Truth, Blue : Ours implemented model
Download the Visual Object Tracking Dataset from website. We have used the data for 2013, 2014 and 2015. More data always helps.
Put the inside folder train_data
as vot2013
, vot2014
and vot2014
. This folder should have folder video wise and inside each folder. extract frames in img
folder.
python runner.py -debug <bool denoting debug on 1 video> -mode <train|test> -rl <bool denoting if RL is on> -model_path <model-path> -vid_path <video-path>
- Loss will be logged at
model_path/loss.log
- You can download our pre-trained model at this link.
Base Code is take from here