Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/dizys/nyu-cv-final-project
NYU CV Final Project: build a AI-generated v.s. real-world-captured image classifier
https://github.com/dizys/nyu-cv-final-project
ai-image-generation dalle2 resnet stable-diffusion
Last synced: 15 days ago
JSON representation
NYU CV Final Project: build a AI-generated v.s. real-world-captured image classifier
- Host: GitHub
- URL: https://github.com/dizys/nyu-cv-final-project
- Owner: dizys
- License: mit
- Created: 2022-11-07T01:38:10.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2022-12-14T18:17:20.000Z (about 2 years ago)
- Last Synced: 2024-12-20T20:03:10.037Z (21 days ago)
- Topics: ai-image-generation, dalle2, resnet, stable-diffusion
- Language: Jupyter Notebook
- Homepage:
- Size: 12.1 MB
- Stars: 0
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# nyu-cv-final-project
NYU CV Final Project: build a AI-generated v.s. real-life-captured image classifier
## Getting Started
### Create virtual environment
```bash
python3 -m venv venv
```
### Activate virtual environment
```bash
source venv/bin/activate
```
### Install dependencies
```bash
pip install -r requirements.txt
```
or when on HPC, to avoid exceeding home inode quota:
```bash
pip install -r requirements.txt --no-cache-dir
# or
pip install -r requirements.txt --no-index --find-links=/scratch/$USER/pip_cache
```
## Get Dataset
We custom built the dataset ourselves from existing public real-photo datasets. We used diffusion-model-based image-to-image models to generate images from real photos, including Stable Diffusion and DALLE2(available only as APIs). This way, we ended up with a labeled dataset of real photos and generated photos.
- How we built the dataset: [dataset/README.md](dataset/README.md)
- Download datasets: [Dataset Releases](https://github.com/dizys/nyu-cv-final-project/releases/tag/dataset)
## Experiment Notebooks
We trained and tested our binary classifier on Google Colab. The notebooks are available in the [notebooks](notebooks) folder.
Training on the main dataset:
- [ResNet34 Train](notebooks/fastai_resnet34_train.ipynb)
- [ResNet50 Train](notebooks/fastai_resnet50_train.ipynb)
Testing on other datasets:
- [ResNet34 Test on Weather Dataset with Stable Diffusion](notebooks/fastai_resnet34_test_weather.ipynb)
- [ResNet50 Test on Weather Dataset with Stable Diffusion](notebooks/fastai_resnet50_test_weather.ipynb)
- [ResNet34 Test on Comic Dataset with Stable Diffusion](notebooks/fastai_resnet34_test_comic.ipynb)
- [ResNet50 Test on Comic Dataset with Stable Diffusion](notebooks/fastai_resnet50_test_comic.ipynb)
- [ResNet34 Test on Small Indoor Dataset with DALLE2](notebooks/fastai_resnet34_test_dalle2_small_indoor.ipynb)
- [ResNet50 Test on Small Indoor Dataset with DALLE2](notebooks/fastai_resnet50_test_dalle2_small_indoor.ipynb)
- [ResNet34 Test on Weather Dataset with DALLE2](notebooks/fastai_resnet34_test_dalle2_weather.ipynb)
- [ResNet50 Test on Weather Dataset with DALLE2](notebooks/fastai_resnet50_test_dalle2_weather.ipynb)
## License
[MIT](LICENSE)