What is the overfitting for human?
To research it , we compare human to deep learning.
main idea : Recognition what clothes own person belong.
Constructing brief-Convolutional Neural Network (CNN).
Number of Convolution layer : 5
Number of Fully Connected : 2
Number of Output node : 5
- Note * Please, change network architecture as appropriate.
Took a picture of T-shirts of each member. Then, these picture use to training dataset.
However, only these are not enough, so that we take a internet.
Capture the clothes you think want to wear on the internet.
About over 10 clothes are collected for each person.
** Training
python3 train.py -i [dataset-path]
We using chainer(v4.0.0).
** Testing
python3 train.py --model [pre-train model path] -i [single-image-path] --test
Please, input single images network, to output display the probability against the input image.
This network is not so deep and number of training image is only a few.
Thus, training is immediately finish and also recognition rate maybe not good.
So when, change the network architecture and increase the dataset size.
python : 3.x.x
PIL : 1.1.7
Pillow : 5.1..0
chainer: 4.x.x
tqdm