xbalazia / active-object-tracking

Active Object Tracking

Download the VIRAT dataset from http://www.viratdata.org/ and take ground truth annotations from the gt.rar archive.

Extract Ground Truth

Extract GT dictionaries with command

$ python3 gt-extract-dict.py -v videos-gt.txt

Find Duplicate Annotations in Ground Truth

Find GT duplicates with command

$ python3 gt-find-duplicates.py -v videos-gt.txt

SSD Detector

Code was originally obtained from https://github.com/elranu/ssd_pi/blob/master/ssd_predictor.py with pre-trained weights from https://github.com/elranu/ssd_pi/blob/master/trained_weights/VGG_coco_SSD_512x512.h5

Run detection with command

$ python3 ssd/ssd.py -v videos-valid.txt

MOSSE Tracker

Code was originally obtained from https://github.com/opencv/opencv/blob/master/samples/python/mosse.py

Run tracking with command

$ python3 mosse/mosse.py videos-valid.txt

VGG16+LSTM+HCF+NMS Regressor

Run training with command

$ ./run-training.sh

Run inference with command

$ ./run-inference.sh videos-valid.txt

Evaluate

Evaluate regression with command

$ python3 eval/eval.py -v vids-valid.txt