simpson_classifier

Classifier for characters in "The Simpsons" (on progress)

Dataset Info

(on progress)

Model info

I trained model with EfficientNet (https://arxiv.org/abs/1905.11946)
and use pre-trained model from geffnet (https://github.com/rwightman/gen-efficientnet-pytorch)
Due to GPU RAM capacity, I use 'tf_efficientnet_b5_ns' model with Image size = 456
I applied data augmentation below using albumentations

alb_train_transform = A.Compose([
    A.Resize(image_size, image_size),
    A.ShiftScaleRotate(shift_limit=0.05,
                       scale_limit=0.05,
                       rotate_limit=15,
                       p=0.5),
    A.HorizontalFlip(),                       
    A.RGBShift(r_shift_limit=15, g_shift_limit=15, b_shift_limit=15, p=0.5),
    A.RandomBrightnessContrast(p=0.5),
    A.GaussNoise(p=0.3),
    A.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
    ToTensorV2()
])

and train model with below settings

batch = 24 
epoch = 20
learning_rate = 1e-5

criterion = nn.CrossEntropyLoss()

optimizer = optim.Adam(params=model.parameters(), lr=learning_rate)

scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer=optimizer,
                                                 T_max=5,
                                                 eta_min=1e-7,
                                                 last_epoch=-1,
                                                 verbose=False)

Result

validation acc: 99.81 Test acc: 99.12

How to

Install the dependencies:

pip install -r requirements.txt

and please run app.py
if you start server with "flask run" torch can't load trained model.

after turning on server, start python interpreter and type below code

import requests

resp = requests.post("http://localhost:5000/predict", files={"file": open('./image_samples/{your_file_name}.jpeg','rb')})

then you can check result at

resp.json()

Available inputs and outputs

Name	Label
Bart simpson	0
charles_monthomery	1
Krusty	2
Lisa_simpson	3
Marge_simpson	4
Milhouse	5
Moe_szyslak	6
Ned_flanders	7
principal_skinner	8

Hesh0629 / simpson_classifier