REQUIREMENTS

• Docker >= 19.03
• CUDA >= 10.0

USAGE

Prepare Videos

Prepare your videos or download from here

Run By Docker Image

Create Docker container

xhost local:root

export VIDEOS=<absolute path of videos folder>
export EXPOSED_PORT=8000

docker run \
	-v $VIDEOS:/mnt/videos \
	-v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=unix$DISPLAY \
	-p $EXPOSED_PORT:8000 \
	--gpus=all -u qtuser -itd \
	qcuong98/ivos-demo

docker exec -it <container_id, output of docker run command> bash

In the container

yarn install && yarn build
./server/run_api.sh

python gui.py \
	[--gpus <gpu_ids for fbrs and stm>] \
	[--mem <memory size>] \
	[--config <json directory>] \
	[--step <step frame>] \
	--video /mnt/videos/<video-name>.mp4

# example: python gui.py --gpus 0 --mem 5 --video /mnt/videos/india.mp4

Annotation results are shown in localhost:EXPORSED_PORT

An example of config file objects:

File objects.json describes that there are 3 object instances in the video sequence. Name of objects with id from 1 to 3 are woman_1, woman_2, and woman_3, respectively.

{
	"objects": [
		"woman_1",
		"woman_2",
		"woman_3",
	]
}

If json config is not specified, name of objects will be object_1, object_2, ..., object_5.

Run By Source

Download pretrained weights

• f-BRS: save in segmentation/fbrs/weights/resnet50_dh128_lvis.pth
• STM: save in propagation/STM/STM_weights.pth

Annotation Tool

python gui.py \
	[--gpus <gpu_ids for fbrs and stm>] \
	[--mem <memory size>] \
	[--config <json directory>] \
	[--step <step frame>] \
	--video <video-path>.mp4

Semantic Video Website

yarn install && yarn build
./server/run_api.sh <port>

CREDIT

A part of this repository is used for DAVIS Challenge 2020 Interactive Scenario

PyQt layout is modified from Seoung Wug Oh's repository