https://github.com/ServiceNow/embedding-propagation
Codebase for Embedding Propagation: Smoother Manifold for Few-Shot Classification. This is a ServiceNow Research project that was started at Element AI.
https://github.com/ServiceNow/embedding-propagation
Last synced: 10 months ago
JSON representation
Codebase for Embedding Propagation: Smoother Manifold for Few-Shot Classification. This is a ServiceNow Research project that was started at Element AI.
- Host: GitHub
- URL: https://github.com/ServiceNow/embedding-propagation
- Owner: ServiceNow
- License: apache-2.0
- Created: 2020-03-09T13:22:59.000Z (almost 6 years ago)
- Default Branch: master
- Last Pushed: 2022-06-27T13:36:41.000Z (over 3 years ago)
- Last Synced: 2024-11-15T07:34:51.387Z (over 1 year ago)
- Language: Python
- Homepage:
- Size: 169 KB
- Stars: 208
- Watchers: 11
- Forks: 21
- Open Issues: 6
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
*ServiceNow completed its acquisition of Element AI on January 8, 2021. All references to Element AI in the materials that are part of this project should refer to ServiceNow.*
[](LICENSE)
Embedding Propagation
Smoother Manifold for Few-Shot Classification [Paper] (ECCV2020)
Embedding propagation can be used to regularize the intermediate features so that generalization performance is improved.
## Usage
Add an embedding propagation layer to your network.
```
pip install git+https://github.com/ElementAI/embedding-propagation
```
```python
import torch
from embedding_propagation import EmbeddingPropagation
ep = EmbeddingPropagation()
features = torch.randn(32, 32)
embeddings = ep(features)
```
## Experiments
Generate the results from the [Paper].
### Install requirements
`pip install -r requirements.txt`
This command installs the [Haven library](https://github.com/haven-ai/haven-ai) which helps in managing the experiments.
### Download the Datasets
* [mini-imagenet](https://github.com/renmengye/few-shot-ssl-public#miniimagenet) ([pre-processing](https://github.com/ElementAI/TADAM/tree/master/datasets))
* [tiered-imagenet](https://github.com/renmengye/few-shot-ssl-public#tieredimagenet)
* [CUB](https://github.com/wyharveychen/CloserLookFewShot/tree/master/filelists/CUB)
If you have the `pkl` version of miniimagenet, you can still use it by setting the dataset name to "episodic_miniimagenet_pkl", in each of the files in `exp_configs`.
### Reproduce the results in the paper
#### 1. Pre-training
```
python3 trainval.py -e pretrain -sb ./logs/pretraining -d
```
where `` is the directory where the data is saved.
#### 2. Fine-tuning
In `exp_configs/finetune_exps.py`, set `"pretrained_weights_root": ./logs/pretraining/`
```
python3 trainval.py -e finetune -sb ./logs/finetuning -d
```
#### 3. SSL experirments with 100 unlabeled
In `exp_configs/ssl_exps.py`, set `"pretrained_weights_root": ./logs/finetuning/`
```
python3 trainval.py -e ssl_large -sb ./logs/ssl/ -d
```
#### 4. SSL experirments with 20-100% unlabeled
In `exp_configs/ssl_exps.py`, set `"pretrained_weights_root": ./logs/finetuning/`
```
python3 trainval.py -e ssl_small -sb ./logs/ssl/ -d
```
### Results
|dataset|model|1-shot|5-shot|
|-------|-----|------|------|
|episodic_cub|conv4|65.94 ± 0.93|78.80 ± 0.64|
|episodic_cub|resnet12|81.32 ± 0.84|91.02 ± 0.44|
|episodic_cub|wrn|87.48 ± 0.68|93.74 ± 0.35|
|episodic_miniimagenet|conv4|57.41 ± 0.85|72.35 ± 0.62|
|episodic_miniimagenet|resnet12|64.82 ± 0.89|80.59 ± 0.64|
|episodic_miniimagenet|wrn|69.92 ± 0.81|83.64 ± 0.54|
|episodic_tiered-imagenet|conv4|58.63 ± 0.92|72.80 ± 0.78|
|episodic_tiered-imagenet|resnet12|75.90 ± 0.90|86.83 ± 0.58|
|episodic_tiered-imagenet|wrn|78.46 ± 0.90|87.46 ± 0.62|
Different from the paper, these results were obtained on a run with fixed hyperparameters during fine-tuning: lr=0.001, alpha=0.2 (now default), train_iters=600, classification_weight=0.1
### Pre-trained weights
https://zenodo.org/record/5552602#.YV2b-UbMKvU
## Citation
```
@article{rodriguez2020embedding,
title={Embedding Propagation: Smoother Manifold for Few-Shot Classification},
author={Pau Rodríguez and Issam Laradji and Alexandre Drouin and Alexandre Lacoste},
year={2020},
journal={arXiv preprint arXiv:2003.04151},
}
```