cyrildiagne/u2net-http

HTTP Wrapper for U^2-Net (Qin et al, Pattern Recognition 2020)

U^2-Net HTTP

This is an HTTP service wrapper of the model: U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection (Qin et al, Pattern Recognition 2020)

The deploy folder contains configuration files for deployment as serverless container with Knative.

Usage:

docker run --rm -p 8080:80 docker.io/cyrildiagne/u2net-http

Test:

curl -F "[email protected]" http://localhost:8080 -o result.png

Development

• Clone this repository: git clone https://github.com/cyrildiagne/u2net-http.git
• Go into the cloned directory: cd u2net-http
• Clone the official U^2-Net repository
• Download the pretrained model u2net.pth
• Put the file inside the U-2-Net/saved_models/u2net/ folder.

Build from source:

Option 1 - Locally with virtualenv

virtualenv venv
venv/bin/activate

pip install torch==0.4.1
pip install -r requirements.txt

python main.py

Option 2 - Using Docker

After you've retrieved the U^2-Net model.

Download Resnet checkpoint

curl https://download.pytorch.org/models/resnet34-333f7ec4.pth -o resnet34-333f7ec4.pth

docker build -t u2net .
docker run --rm -p 8080:80 u2net

u2net-http's Issues

Why resizing twice during inference?

Hi @cyrildiagne !

Why did you have to ensure the input image size to be under 1024 max dimension (https://github.com/cyrildiagne/u2net-http/blob/master/main.py#L41-L43), when the image will be eventually resized to 320 x 320 in after RescaleT?

Deploying

Sorry for this question, but I read numerous docs on Knative and so on for days, but I cant seem to work out how to deploy this :D

I would like to deploy it to Google Cloud, how exactly do I do this?
Sorry again,
Altay Akkus

RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False.

Hey! First of all, thank you very much for this project I find it very useful. I've tried basnet-http and it worked perfectly but when I try to run this project on OSX I'm getting this error:

RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location='cpu' to map your storages to the CPU.

Do you know if it's possible to fix it?

Image Permission Denied

I am trying to run it with Docker but encountered a permission error:
This issue in basnet-http looks similar

~$ docker run --rm -p 8080:80 gcr.io/cyrildiagne-ml/u2net-http
Unable to find image 'gcr.io/cyrildiagne-ml/u2net-http:latest' locally
docker: Error response from daemon: unauthorized: You don't have the needed permissions to perform this operation, and you may have invalid credentials. To authenticate your request, follow the steps in: https://cloud.google.com/container-registry/docs/advanced-authentication.
See 'docker run --help'.

U2Net is accepted by Pattern Recognition 2020 not CVPR 2020.

Hi, Cyril,

I am the author of BASNet and U2Net. You AR CUT and PASTE project is fantastic. We are very glad that our model is used in your project. Just one thing to mention that the U^2Net paper is accepted in Patter Recognition 2020 not CVPR 2020.

Best wishes,
Xuebin