initial commit

.gitignore

@@ -0,0 +1,3 @@
+# PhpStorm / IDEA
+.idea
+

README.md

@@ -1,3 +1,30 @@
 ---
-license: mit
+tags:
+  - vision
+  - image-to-image
+  - endpoints-template
+inference: false
+pipeline_tag: image-to-image
+base_model: timbrooks/instruct-pix2pix
+library_name: generic
 ---
+
+# Fork of [timbrooks/instruct-pix2pix](https://huggingface.co/timbrooks/instruct-pix2pix) for an `image-to-image` Inference endpoint.
+
+This repository implements a `custom` task for `image-to-image` with instructions for 🤗 Inference Endpoints. The code for the customized 
+pipeline is in the handler.py.
+
+To use deploy this model an Inference Endpoint you have to select `Custom` as task to use the `handler.py` file. 
+
+### expected Request payload
+
+```json
+{
+  "image": encoded_image,
+  "parameters": {
+    "candidate_labels": "green, yellow, blue, white, silver"
+  }
+}
+```
+
+`encoded_image` is a base64 encoded image. 

handler.py

@@ -0,0 +1,34 @@
+from typing import Dict, List, Any
+from PIL import Image
+from io import BytesIO
+import torch
+import base64
+from diffusers import StableDiffusionInstructPix2PixPipeline, EulerAncestralDiscreteScheduler
+
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+class EndpointHandler():
+    def __init__(self, path=""):
+        model_id = "timbrooks/instruct-pix2pix"
+        self.pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(model_id, torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32, safety_checker=None)
+        self.pipe.to(device)
+        self.pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(self.pipe.scheduler.config)
+
+    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
+        """
+       data args:
+            inputs (:obj:`string`)
+            parameters (:obj:)
+      Return:
+            A :obj:`string`:. Base64 encoded image string
+        """
+
+        image_data = data.pop("inputs", data)
+        # decode base64 image to PIL
+        image = Image.open(BytesIO(base64.b64decode(image_data)))
+
+        parameters = data.pop("parameters", data)
+        prompt = parameters['prompt']
+
+        images = pipe(prompt, image=image, num_inference_steps=10, image_guidance_scale=1).images
+        return images[0]