ImageCNN

Dataset

Download & unzip pictures under data directory

wget http://download.tensorflow.org/example_images/flower_photos.tgz

Examples


daisy	dandelion	roses	sunflowers	tulips

Split files into two folders: train and test with tools/shuflink cd data
../tools/shuflink flower_photos train test
cd ..

Train

$python train.py -h

Training models are saved under logs. After training, you should move one of them to model directory and rename it as “flower.model“ for evaluation and prediction.

Install visdom

pip install visdom

start visdom

python -m visdom.server &

Start Train

python train.py

Loss and Acc

http://127.0.0.1:8097

Evaluate

$ python eval.py -h

$ python eval.py

Evaluating data information:
Dataset ImageFolder
Number of datapoints: 918
Root Location: data/test
Transforms (if any): Compose(
Resize(size=256, interpolation=PIL.Image.BILINEAR)
CenterCrop(size=(224, 224))
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
Target Transforms (if any): None
Class names: ['daisy', 'dandelion', 'roses', 'sunflowers', 'tulips']
2018-09-28 15:31:19,234 - model.py[line:203] - INFO: Start evaluating ...
2018-09-28 15:31:21,553 - model.py[line:222] - INFO: Evaluating ACC: 99.24%
2018-09-28 15:31:21,553 - model.py[line:223] - INFO: Evaluating finished.

Predict

python predict.py -h

python predict.py images/daisy_5794835_d15905c7c8_n.jpg

Image class: 0, daisy, 1.00, images/daisy_5794835_d15905c7c8_n.jpg

Requirements

Python 3.6
Pytorch 0.4.0, 1.0
Torchvision 0.2.1
Pillow 4.2.1

License

ImageCNN is released under the [Apache 2.0 license]

Chinese Document

通用图像分类器.

delldu/ImageCNN