Skip to content

Latest commit

 

History

History
55 lines (25 loc) · 1.51 KB

readme_deepsort.md

File metadata and controls

55 lines (25 loc) · 1.51 KB
Raw

Deepsort with RetinaNet

Deepsort forked from : https://github.com/nwojke/deep_sort

Requirements :

  • python = 3.6

  • Tensorflow >= 1.12

  • numpy >= 1.14.5

  • cv2 >= 3.4.1

Inputs :

  • Video

  • Trained RetinaNet model

  • Trained deepsort model

Args :

-m , --model : Path to the trained RetinaNet model

-tm , --tracker_model : Path to tensorflow frozen graph of tracker (deepsort) model

-c , --confidence : Detection threshold for the RetinaNet model . default = 0.8

-v , --video : Path to input video file

-f , --fps : FPS for output video . default=25

-s , --skip : Sampling interval for the inference to run . default=0 ie., no skipping of frames

-sc , --scale : Scaling for the input frame . default=1 ie., no scaling

-age , --max_age : Max frames to wait before discarding a vanished ID . default=100

-n_init , --n_init : Number of frames that a track remains in initialization phase . default=1

Sample Args :

For simple deepsort :

python run.py -v video.mov -m snapshots/version8_resplit_test_train/resnet50_csv_12_inference.h5 -tm deep_sort/resources/networks/mars-small128.pb -n_init 5 -age 100 -c 0.5 -s 0 -sc 0.5