CenterNET_4347

Testing for Machine Learning project.

CenterNet Deep SORT

-This contains the basic instructions for installation of centernet deepsort. After this has been installed change the filepaths on the centernet_deepsort2PROJ.py file to meet the requirements to produce a bounded boxed video. A demo video is included with the direct submission.

-the necessary MODEL.pth file and video test file unboxed will be included in the final deliverable.

Requirements / Installation

git clone https://github.com/kimyoon-young/centerNet-deep-sort.git
cd centerNet-deep-sort
conda env create -f CenterNet.yml
pip install -r requirments.txt

Training CenterNet

• Follow installation instructions
• Basic training command
• To re-train an existing model (e.g. COCO-DLA), use the option --load_model [path_to_model]
• CUDAoutofmemory exceptions may occur; start with batch size of 16x(numGPUs) and adjust accordingly.

Example command for training on the RedBarn workstation in M12

python main.py ctdet --exp_id coco_dla --batch_size 16 --master_batch 15 --lr 1.25e-4  --load_model /centerNet-deep-sort/CenterNet/models/ctdet_coco_dla_2x.pth --gpus 0

For a COCO model that has already been retrained on IPATCH, visit https://webfiles.txstate.edu/ -> DepartmentShare/AA/COSE/CS/CS-Shares/BigDataM12/NAVAIR/CenterNetDeepSORT/model_best.pth (TXState Credentials Required)

Evaluating Model

Follow the benchmark evaluation instructions, but use the model you want to evaluate rather than one from their model zoo.

CNDS Tools

Used for extracting tracking data from a directory of videos and annotation files.