SSVD

Implementation of "Single Shot Video Object Detector". The original version is implemented with MXNet, but we convert it into pytorch and take the off-the-shelf library from maskrcnn-benchmark to make it easier to try.

Installation

See INSTALL.md

Start

1. Dataset preparation

Download the ILSVRC DET and ILSVRC VID datasets, unzip and move them to "./dataset". The structure of dataset looks like:

./dataset/ILSVRC/
./dataset/ILSVRC/Annotations/DET
./dataset/ILSVRC/Annotations/VID
./dataset/ILSVRC/Data/DET
./dataset/ILSVRC/Data/VID
./dataset/ILSVRC/ImageSets

2. Model downloading

Download our model from https://drive.google.com/file/d/1vJkGGgCO-NmzzJmV5vwFisw_3D8VJ_L4/view?usp=sharing.

Create a folder named "ckpt", and put our model into it.

3. Setup

python setup.py build develop

4.Test on ILSVRC VID dataset

bash ssvd_test.sh