https://github.com/gs18113/ESPCN-TensorFlow2

ESPCN(https://arxiv.org/abs/1609.05158) implementation using TensorFlow 2.0
https://github.com/gs18113/ESPCN-TensorFlow2

Last synced: 2 months ago
JSON representation

ESPCN(https://arxiv.org/abs/1609.05158) implementation using TensorFlow 2.0

• Host: GitHub
• URL: https://github.com/gs18113/ESPCN-TensorFlow2
• Owner: gs18113
• Created: 2019-10-23T14:08:02.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2019-11-04T02:56:18.000Z (about 5 years ago)
• Last Synced: 2024-08-02T19:01:49.392Z (6 months ago)
• Language: Jupyter Notebook
• Size: 2.04 MB
• Stars: 4
• Watchers: 3
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Tensorflow2 - gs18113/ESPCN-TensorFlow2

README

        
# ESPCN-TensorFlow2

ESPCN(https://arxiv.org/abs/1609.05158) implementation using TensorFlow 2.0

## Requirements

* tensorflow >= 2.0.0

* tensorflow_datasets

## Note

* **The model architecture is slightly different from the paper**.(Especially filter number)

* The phase-shifting code is from https://github.com/kweisamx/TensorFlow-ESPCN.

* The COCO dataset by tensorflow_datsets takes up really much space! (About 77G in my case...)

* TPU code is still in development.

* This code was written for studying, so the code may be hard to understand... I'll try my best to improve code readability.

## Usage

### Train

```

python3 train.py -exp_name EXP_NAME [-lr LEARNING_RATE] [-batch_size BATCH_SIZE] [-save_dir SAVE_DIR]

```

Learning rate about 0.001 is recommended. The model won't converge correctly if lr is too large.

### Convert to TFLite

```

python3 tflite.py -exp_name EXP_NAME -model_epoch EPOCH [-saved_dir SAVED_DIR] [-tflite_dir TFLITE_DIR]

```

I recommend reading the code for more training / converting options.