SiameseCAR demo

1. Structure

|-- siamcar  

    |-- core
       |-- config.py  
       |-- xcoor.py  

    |-- model
        |-- backbone.py
        |-- head.py 
        |-- neck.py  
        |-- model_builder.py 
        |-- loss_car.py 
        |-- loss_cls.py 
  
    |-- tracker 
        |-- base_tracker.py 
        |-- siamcar_tracker.py 
    
    |-- utils
        |-- car_utils.py
        |-- iou_loss.py
        |-- model_load.py  
 
    |-- config.yaml
    
|-- testing_dataset 
 
    *put the test video in this file*
 
|-- tools
    |-- snapshot *put the pretrained model in this file*  
    |-- demp.py

2. Environment setup
This code has been tested on Ubuntu 16.04, Python 3.6, Pytorch 0.4.1 & 1.2, CUDA 9.0.

3. Test date
Put the testing videos or picture sequences into testing_dataset directory. During the testing, you need to mark the tracking target in the first frame with a bounding box.

4. Tracker
Download the pretrained model and put them into tools/snapshot directory.
general_model
got10k_model
LaSOT_model

5. Testing demo

cd SiamCAR/tools python demo.py \
--video
--snapshot
--hp_search

Optional parameters:

• --video:path of the testing video or picture sequences(eg:--video testing_dataset/people.mp4 or --video testing_dataset/Biker)
• --snapshot:path of the specified testing model. We provide three model for testing. The general model was trained in VID, YOUTUBEBB, COCO and DET. The LaSOT model was only trained in LaSOT training dataset, and the got10k model was only trained in GOT10K training dataset. (eg:--snapshot ./snapshot/general_model.pth or --snapshot ./snapshot/LaSOT_model.pth or --snapshot ./snapshot/got10k_model.pth)
• hp_search:There are different hyper paramters for different datasets. (eg:--hp_search OTB/LaSOT/VOT2019/GOT10k/UAV123)

6. Acknowledgement
The code is implemented based on pysot. We would like to express our sincere thanks to the contributors.