Crosslink-NX VGG-SqDet Human-Count

A network backbone with VGG and SqueezeDet detector to Power Human Counting usecase.

Dataset

Note that the dataset should be in KITTI format. Images and labels should be arranged as follows:

• Training Images: 'data_path'/training/images
• Labels: 'data_path'/training/labels
• train.txt:'data_path'/ImageSets/train.txt
• Test Images: 'data_path'/test/img
• Test Labels: 'data_path'/test/labels

Setup Environment

$ pip install -r requirements.txt

Run Training

1. Configure scripts/train.sh

   • Mainly configure dataset path variable: TRAIN_DATA_DIR
   • Optionally User can change other configurations.

2. Run command :

   $ ./run

   Freezing Model

   $ python src/genpb.py –ckpt_dir <log directory> --freeze