erdiaker / resnet

A deep residual network implementation using TF high-level APIs.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ResNet

An implementation of Deep Residual Learning for Image Recognition using Tensorflow high-level APIs.

Data

Scripts to train and evaluate the model with Open Images V4 dataset are included in data.py. Instructions to download the dataset are here.

Dependencies

pip3 install -r requirements.txt

Training and Evaluation

python3 train.py \
    --train_tar_path='train.tar.gz' \ 
    --train_annotation_path='train-annotations-human-imagelabels-boxable.csv' \
    --valid_tar_path='validation.tar.gz' \
    --valid_annotation_path='validation-annotations-human-imagelabels-boxable.csv' \
    --depth=50 \
    --model_dir='./logs'

Type python3 train.py --help to see the full list of arguments.

Prediction

python3 predict.py \
    --test_tar_path='test.tar.gz' \ 
    --depth=50 \
    --model_dir='./logs' \
    --output_type='labels'

Type python3 predict.py --help to see the full list of arguments.

About

A deep residual network implementation using TF high-level APIs.


Languages

Language:Python 100.0%