/A-fall-detection-model-in-diverse-conditions

Primary LanguagePython

AGC_final

download weights

You can download the pre-trained weight files in this link, and put it in the AGC_final directory.

environments

torch==1.8.1
opencv
albumentations
timm
python-bidi
soynlp
pandas
matplotlib
seaborn

inference code

  1. download the pre-trained .pth weight files in the save_models. There are many pre-trained models for running the code.
  2. run the prediction code, and specify your own video dataset directory with pre-defined.json. We provide the sample video.mp4 and pre-defined.json files.
python main.py --video_root_dir sample_dataset
  1. You can see the predicted json files in the ./debug/answer.
  2. You can also see the first ocr prediction results in the ./debug/first_ocr_results

first frame ocr results

image image

fallen person detection, classification and indicating where the perfon falls.

image

fallen_low.mp4