Deep Learning Frameworks Performance Evaluation

This project is targeted to evaluate the performances of three deep learning frameworks on some different and typical deep nerual networks, including:

1. Google TensorFlow
2. Facebook PyTorch
3. Baidu PaddlePaddle

These three frameworks are evaluated by the following tasks:

• task1: image classification on CIFAR10 by VGG16
• task2: text sentimental classification on IMDB movie reviews by bi-lstm
• task3: text sentimental classification on IMDB movie reviews by bert
• task4: image generation on MINST dataset by DCGAN
• task5: image translation on horse2zebra dataset by CycleGAN (Paddle implementation not included)

Project Structure

All tasks implemented by these three frameworks are placed in three directories named by the framework. Three directories are:

1. Tensorflow
2. Pytorch
3. Paddle

In each framework's directory, each task is implemented in directoy as:

• task1 in directory: imageclassification
• task2 in directory: textclassify_bilstm
• task3 in directory: textclassify_bert
• task4 in directory: DCGAN
• task5 in directory: CycleGAN

Run the task

Please refer the readme in each framework's directory