CADC-SVHN-PyTorch

Model designed to recognize a number with two digits, trained on SVHN dataset, implemented using PyTorch. Automatic parameter searching using NNI.

Requirements

• torch 1.0.1

• torchvision 0.2.1

• Pillow

• protobuf

• lmdb

• h5py

  $ sudo apt-get install libhdf5-dev
  $ pip install h5py
  

• NNI (for AutoML)

Setup

2. Download SVHN Dataset format 1

3. Extract to data folder, now your folder structure should be like below:

   SVHNClassifier
       - data
           - extra
               - 1.png 
               - 2.png
               - ...
               - digitStruct.mat
           - test
               - 1.png 
               - 2.png
               - ...
               - digitStruct.mat
           - train
               - 1.png 
               - 2.png
               - ...
               - digitStruct.mat
   

Data pre-process

1. Convert to LMDB format

   $ python convert_to_lmdb.py --data_dir ./data
   

   Please note that I have cut picture to 2 digits, resize them to 5454 and suppressed all pictures small than 5454, you may make changes in order to meet your requirement.

2. (Optional) Test for reading LMDBs

   Open read_lmdb_sample.ipynb in Jupyter

Download the converted and processed lmdb dataset

链接: https://pan.baidu.com/s/1Q7d1BrFzn7l5eLAnD8ckyQ 密码: owdc

Train, test and convert to .pt

Open `mobilenet.ipynb` in Jupyter

AutoML

1. (Optional) modify parameters in search_space.json and in config.yml
2. $ nnictl create --config ./config.yml