Goal is to filter out inappropriate pictures (like a SIM image).
An image is considered to be inappropriate if a human cannot evaluate the parity by looking at it
Used parts of Chicago dataset (306 images)
Classes:
- Good
- Trash
For this task we used pre-trained (on imageNet) ResNet50, and made fine-tuning on the custom dataset.
Combined 2 datasets: 296297_project.tar (870 images) and 296300_project.tar (535 images)
Classes:
- Megafon
- Yota
- Opponents
Augmentation applied:
- Flip: Horizontal, Vertical
- 90° Rotate: Clockwise, Counter-Clockwise
- Crop: 0% Minimum Zoom, 20% Maximum Zoom
- Blur: Up to 2.5px
- Noise: Up to 1.2% of pixels
About:
Validation predicts and labels comparison
Metrics and losses:
About:
Validation predicts:
