U-2-Net-Demo
Demonstration using Google Colab to show how U-2-NET can be used for Background Removal, Changing Backgrounds, Bounding Box Creation, Salient Feature Highlighting and Salient Object Cropping
Please let me know if you would like to collaborate on building a website/web demo for this repo
Check out these articles making use of this repository:
Medium: Large Scale Background Removal Using AI
I REMOVED THE BACKGROUNDS FROM 11,782 JPGS FOR FREE! USING AI!
UPDATES: upcoming update - notebook for inferencing a large number of images
- [Jan 11, 2021] - Salient Object Cropping notebook is now available following this PR, thanks to the contribution from Norod!
- [Jan 05, 2021] - Keras .h5 saved models available in new repository following this issue Note: if there are any new requests for a TensorFlow saved model, I will upload
- [Dec 24, 2020] - change background in step by step file: commit
- [Oct 29, 2020] - implemented quick fix for ipynb.checkpoints error: commit
Link: Open Google Colab Notebook
Click this link for step-by-step instructions: Open Google Colab Notebook {Use this to save your results too} {update: now with code to change background}
Click this link for Salient Object Cropping: U_2_Netp_Cropper_Colab
Example Results
Cat:
Formula 1 Car:
Rick Astley:
Background Changed:
Bacground Removed and Image Cropped (Mask and Resulting Image):
Useful material
U-2-NET Paper: U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection
Original Repo: U-2-Net Github repo
Modified repo for that this colab uses: Modified fork
References: X. Qin, Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, and M. Jagersand, “U2-net: Going deeper with nested u-structure for salient object detection,” Pattern Recognition, vol. 106, p. 107404, 2020
Other Results
Details about U-2-Net
The following is an excerpt from the paper: 'In this paper, we design a simple yet powerful deep network architecture, U2-Net, for salient object detection (SOD). The architecture of our U2-Net is a two-level nested U-structure. The design has the following advantages: (1) it is able to capture more contextual information from different scales thanks to the mixture of receptive fields of different sizes in our proposed ReSidual U-blocks (RSU), (2) it increases the depth of the whole architecture without significantly increasing the computational cost because of the pooling operations used in these RSU blocks. This architecture enables us to train a deep network from scratch without using backbones from image classification tasks. We instantiate two models of the proposed architecture, U2-Net (176.3 MB, 30 FPS on GTX 1080Ti GPU) and U2-Net† (4.7 MB, 40 FPS), to facilitate the usage in different environments. Both models achieve competitive performance on six SOD datasets.'
U2NET Model
TODO:
- remove dependancy to secondary repo
- algorithmic post processing
- flask demo
- multiproc cpu code
- benchmarks
- video background removal
- change backgrounds for main notebook
- support .jpeg images
- upload python code for webcam support
- add demo for portrait feature of U-2-Net
- add license to nbs
- maybe make a webpage for bounding box creation over large datasets
Please feel free to suggest changes to improve this repository