pix2pix-http

pix2pix-http

pix2pix as an http service.

Feed and output image and generate an output image using different models over http requests.

Requirements

• TensorFlow (>= 1.4.1)
• pix2pix-tensorflow
• Set the PIX2PIX_PROCESSLOCALPY environment variable (in the .env file) path to your pix2pix process-local.py path (i.e., pix2pix-tensorflow/server/tools/process-local.py).

Usage

Run the http server at 127.0.0.1.

go run *.go

The server provides two POST routes.

• /pix · POST Route that gets an input image_file and a model_path and retrieves a generated pix2pix output image using the pix2pix model at model_path and the provided image_file.
  • @param image_file · a PNG input image file of 256x256 pixels.
  • @param model_path · the path to a pix2pix model_export/ exported model.
  • @output image · bytes of the PNG output image (also 256x256 pixels).
• /blur · (Test) POST route that gets an input image and retrieves it blurred with the provided sigma.

Sample POST curl command

make pix path=/path/to/models/170331_pix-08-edges2sunflowers-200e/model_export

Makefile

# Generate an output image with a pix2pix model path (or with a model id and models root path).
# (a) providing the model_export path of your pix2pix model
# e.g., make pix path=3 path=/path/to/your/model_export
# (b) providing the name of the model and the root directory where your models live
# e.g., make pix model=3 dir=/path/to/your/models
# e.g., make pix model=3 name=tulips
pix:
	@curl -X POST --form "images_file=@flower.png" \
	--form "model_path=$(path)" \
	--form "model=$(model)" \
	--form "dir=$(dir)" \
    "http://localhost:8000/pix2pix" > "flower-pix2pix-$(name).png"
	@echo ""

# Blur an image.
# e.g., make blur sigma=10
blur:
	@curl -X POST --form "images_file=@flower.png" \
	--form "sigma=$(sigma)" \
	"http://localhost:8000/blur" > "flower-blur.png"
	@echo ""