Fisheye hand localizer
When there is a fisheye lense distorted image data, how does one resolve the issue of localisation? So I have posed a particular problem of attempting to localize the hand from a fisheye distorted image data.
Result and comparison
Refer to the notebook in the directory.
Files
- localizer.py
fisheye localizer script involves hard negative mining
- preprocess.py
generates cropped hand in ./dataset/cropped_hand and a csv file in cropped_hand directory | it makes use of the orignal dataset | includes the side of the cropped hand
- pretrained_undistorted.py
Using pretrained model, crop the hand from an image and save cropped image to ./result/...
- pretrained_undistorted.py
Using pretrained model, crop the hand from an image and save cropped image to ./dataset/cropped_hand
- fisheye_position.py
Generates a csv file in fish_hand directory | csv includes the location of vertices of distorted image
- util.py
Etc functions required for test and modeling
- gesture_recognizer1.py
Generates plain handDetector.pkl at current directory
./data_augment/*
Files for generating augmented data