This repository contains codes and pre-trained checkpoints of a place recognition model for Friends video.

• Ubuntu (tested on v14.04)
• Python (tested on v2.7)
• Pytorch (tested on v0.4.1)
• Numpy
• PIL
• opencv-python
• matplotlib
• jsonl

Pre-trained checkpoint should be placed in the root directory.

You can download the pre-trained checkpoint from this link

Input: video file (e.g. *.avi, *.mkv)

Output: jsonl file (predicted class of video frames for every second, 1 fps)

Below is an example of output jsonl flle.

{"second": 0.0, "type": "location", "class": "none"}
{"second": 0.0, "type": "location", "class": "none"}
...
{"second": 52.0, "type": "location", "class": "cafe"}
{"second": 53.0, "type": "location", "class": "cafe"}
...
{"second": 314.0, "type": "location", "class": "home-livingroom-Monica"}
{"second": 315.0, "type": "location", "class": "home-livingroom-Monica"}
...

You can run following command on a terminal,

python demo.py <video-file> <output-file-name>

For example

python demo.py input.avi output

Then output.jsonl file will be saved in the root directory

This project was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIT) (2017-0-01780, The technology development for event recognition/relational reasoning and learning knowledge based system for video understanding)