Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- [Dataset Preparation]
- [Step-1]:
- Input image should be randomly croped & resized to 12 x 12 as PNet process small size
- image at a faster speed to propose face regions.
- [Step-2]:
- Iou(croped_image, ground_truth) is calculated.For negative image : Iou(croped_image, ground_truth) <= 0.3
- positive image(where Face is present) : Iou(croped_image, ground_truth) >= 0.6
- Ignore Image/part image : Iou(croped_image, ground_truth) <=0.4.
- [Step-3]:
- This pos, neg & ignore samples are saved in tfrecord format & are used for training PNet.
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement