Compressed ESRGAN Tensorflow Hub and TFLite Model (DIV2K)

Compressed ESRGAN for TensorFlow Hub

Compressed Enhanced Super Resolution GAN for image super resolution.

import tensorflow_hub as hub
import tensorflow as tf
model = hub.load("https://github.com/captain-pool/GSOC/releases/download/"
                 "2.0.0/compressed_esrgan.tar.gz")
super_resolution = model.call(LOW_RESOLUTION_IMAGE_OF_SHAPE=[BATCH, 180, 320, 3])

Results

Latency: 0.4 Seconds per image
Mean PSNR: 25.3

NOTE

• The image must be a float32 image, converted using tf.cast(image, tf.float32).
• To display the image, don't forget to convert back to uint8 using
  tf.cast(tf.clip_by_value(image, 0, 255), tf.uint8)

The Optimized TFLite file is also attached.

Input Shape: [1, 180, 320, 3]
Output Shape: [1, 720, 1280, 3]

Enhanced Super Resolution GAN Tensorflow Hub Model

ESRGAN for Tensorflow Hub

Enhanced Super Resolution GAN for image super resolution.
Produce x4 Super Resolution Image from images of Height, Width >=64.

import tensorflow_hub as hub
import tensorflow as tf
model = hub.load("https://github.com/captain-pool/GSOC/releases/download/1.0.0/esrgan.tar.gz")
super_resolution = model(LOW_RESOLUTION_IMAGE_OF_SHAPE=[BATCH, HEIGHT, WIDTH, 3])

NOTE

• The image must be a float32 image, converted using tf.cast(image, tf.float32).
• To display the image, don't forget to convert back to uint8 using
  tf.cast(tf.clip_by_value(image, 0, 255), tf.uint8)