https://github.com/kshitijd20/RSA-CVPR19-release
code for CVPR paper "Representation Similarity Analysis for Efficient Task Taxonomy and Transfer Learning"
https://github.com/kshitijd20/RSA-CVPR19-release
Last synced: 7 months ago
JSON representation
code for CVPR paper "Representation Similarity Analysis for Efficient Task Taxonomy and Transfer Learning"
- Host: GitHub
- URL: https://github.com/kshitijd20/RSA-CVPR19-release
- Owner: kshitijd20
- Created: 2019-04-03T04:41:11.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2022-12-08T01:04:52.000Z (almost 3 years ago)
- Last Synced: 2024-11-15T06:33:21.384Z (about 1 year ago)
- Language: Python
- Size: 17.9 MB
- Stars: 26
- Watchers: 6
- Forks: 8
- Open Issues: 9
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- Awesome-Multi-Task-Learning - [PyTorch
README
* ***News!!! We introduced a new generic framework to measure similarity between tasks. Please find our [code](https://github.com/cvai-repo/duality-diagram-similarity) and [paper(accepted in ECCV 2020)](https://arxiv.org/abs/2008.02107).
# CVPR_RSA_Taskonomy
code for CVPR paper ["Representation Similarity Analysis for Efficient Task Taxonomy and Transfer Learning"](https://arxiv.org/abs/1904.11740)
This repo is build upon the code from the [Taskonomy paper](http://taskonomy.stanford.edu/).
#### Step1 RDM creation
1. Switch to taskonomy/taskbank
2. Follow installation instructions and downloading pretrained models
3. Download taskonomy images used for RDM creation [here](https://drive.google.com/open?id=1IwP1_lfLl88FXq0AvfmYWUtw1Z5e6urT)
4. Change current directory to taskonomy/taskbank/tools directory
5. Run python generate_features_taskonomy_500 --img_dir "path to downloaded image directory" --save_dir "path to save features"
6. Run python create_taskonomy_RDMs_enc_final.py --feat_dir "path to saved features" --save_dir "path to save RDMs"
P.S.: Make sure you have enough space as saved features require lot of space (67 GB).
#### Step2 Download my forked copy of [RSA toolbox](https://github.com/kshitijd20/rsatoolbox)
#### Step3 RSA analysis (generate figures from the paper)
1. Change current directory to RSA_analysis
2. run demo.m after modifying the rdm_dir, rsa_dir, and save_dir paths to corresponding paths in the demo.m to generate results for Figure 4 and 5 from the paper
The code to generate results for small models and pascal VOC will be released soon. For early access or any other queries please contact kshitijdwivedi93@gmail.com