This repository includes scripts to make training HoG feature classifier on OpenCV easier.
- Python
- libsvm
- numpy
- OpenCV (must be exposed to Python interface)
Change necessary variables in train.sh and execute the file.
| Variable | Description |
|---|---|
| POS_PATH | Path to positive training set images |
| NEG_PATH | Path to negative training set images |
| OUTPUT_PATH | Path to save resulting files |
| MODEL_NAME | Name to save the resulting files |
| WIDTH | Sliding window width |
| HEIGHT | Sliding window height |
| TRAINER_PATH | Path to this folder |
| Filename | Description |
|---|---|
| modelname | File containing extracted HoG features, in libsvm format |
| modelname.model | File containing results form libsvm format |
| modelname.features | File containing hyperparameter detecting vectors |
| modelname.features is the only required file in runtime. |
Currently, the width/height value must follow the following rule:
(WIDTH - 16) % 8 = 0
(HEIGHT - 16) % 8 = 0
This will be changed in the future.