conda create --name <yourenvname> python=3.7.4
conda activate <yourenvname>
conda install --yes --file requirements.txt
kaggle package is not available in conda so you need to install it with pip, however conda env supports pip command
pip install kaggle==1.5.6
For serving with GPU set environment variable
$ export TF_FORCE_GPU_ALLOW_GROWTH=true
$ sh data_download.sh
cd backend/src
python model.py
Note: Everytime you run model.py, it starts training from latest checkpoint. Only the latest model weights will be saved.
$ tensorboard --logdir=temp
cd backend
python app.py
cd frontend
npm i && npm start
Input | Input Size | Kernel | Stride | Num Kernels | Output Size |
---|---|---|---|---|---|
Image | 150* 150 *3 | 9 * 9 * 3 | 3 | 64 | 48 *48 *64 |
Max Pool | 48* 48 *64 | 2 * 2 | 2 | -- | 24 * 24 * 64 |
Conv 1 | 24 * 24 *64 | 5 * 5 * 64 | 1 | 32 | 20 * 20 *32 |
Max Pool 1 | 20* 20* 32 | 3 * 3 | 1 | -- | 18 * 18 * 32 |
Conv 2 | 18 * 18 * 32 | 3 * 3 * 32 | 1 | 16 | 16 * 16 * 16 |
Max Pool 2 | 16 * 16 *16 | 3 * 3 | 1 | -- | 14 * 14 * 16 |
Input Layer | Input Shape | Output Shape | Output Layer |
---|---|---|---|
Max Pool 2 | 14 * 14 * 16 | 3136 | Flatten |
Flatten | 3136 | 512 | Dense 1 |
Dense 1 | 512 | 256 | Dense 2 |
Dense 2 | 256 | 64 | Dense 3 |
Dense 3 | 64 | 6 | Final Output |