This repo will adress the following aspects of using AWS for training deep learning models
- General setup for AWS EC2 (Elastic Compute Cloud) and S3 Storage Classes for training deep learning model using Pytorch
- The implementation for paper Deep Fashion Analysis with Feature Map Upsampling and Landmark-driven Attention (2018) is taken as example to show the workflow
- Manage the remote terminal sessions with tmux.
- Data preprocessing for duplicated attributes.
- Load the GPU-trained model and run on local machine.
After following the post's creating info.csv step, it is necessary to do some data cleaning since the original DeepFashion Dataset has a lot duplicated labels, such as : stripe, stripes, striped are listed as separate attributes.
The demostration for running the scripts are inside the data_cleaning file.
Using preprocessed data for training could significantly improve the accuracy, the results are follows:
The following table shows the category classification and attribute prediction results on the DeepFashion dataset for the original dataset and our pre-processed dataset. The two numbers in each cell stands for top-3 and top-5 accuracy.
| Methods | Category | Texture | Fabric | Shape | Part | Style | All |
|---|---|---|---|---|---|---|---|
| Liu.et al. | 91.16 | 96.12 | 56.17 | 65.83 | 43.20 | 53.52 | 58.28 | 67.80 | 46.97 | 57.42 | 68.82 | 74.13 | 54.69 | 63.74 |
| Ours | 90.99 | 95.88 | 69.92 | 78.86 | 50.42 | 60.91 | 64.02 | 72.43 | 59.03 | 68.87 | 37.42 | 46.40 | 31.13 | 46.40 |
The workflow for this specific example will provide a standard practice for loading trained model and put it into use on your local machine. the code is in file Model-Implementation.
The demostration for using the classification model is in the readme.md inside Model-Implementation file.