Flask app and tg bot for food classification
First, load the models https://drive.google.com/drive/folders/1P1h7bcjNZuFKtavjYokkzPebYxdHpE56?usp=sharing
Build application
$ https://github.com/Argentum471/food_classification.git
$ docker-compose up
Bot: https://t.me/GastronomyGeniusBot
If you have .env, put it in root
Reuslts
model | top1 |
---|---|
resnet18 | 0.73 |
resnet50d | 0.82 |
tf_efficientnetv2_s.in1k | 0.88 |
Training parameters
Parameter | Value |
---|---|
epochs | 100 |
optimizer | Adam |
scheduler | ReduceLROnPlateau |
lr | 3e-4 |
How to improve
- more epochs
- more data for worst classes
- new architecture
- more specified augmentations