Ubuntu
commited on
Commit
·
c2fbc3b
1
Parent(s):
5d0eb5a
add custom handler
Browse files- handler.py +48 -0
- requirements.txt +6 -0
- test_handler.py +13 -0
handler.py
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import Dict, List, Any
|
| 2 |
+
import os
|
| 3 |
+
import requests
|
| 4 |
+
from flask import Flask, Response, request, jsonify
|
| 5 |
+
from segment_anything import SamPredictor, sam_model_registry
|
| 6 |
+
|
| 7 |
+
class EndpointHandler():
|
| 8 |
+
def __init__(self, path=""):
|
| 9 |
+
# Preload all the elements you are going to need at inference.
|
| 10 |
+
model_type = "vit_b"
|
| 11 |
+
# prefix = "/opt/ml/model"
|
| 12 |
+
model_path = "tf_model.h5"
|
| 13 |
+
# model_checkpoint_path = os.path.join(prefix, "sam_vit_h_4b8939.pth")
|
| 14 |
+
|
| 15 |
+
sam = sam_model_registry[model_type](checkpoint=model_path)
|
| 16 |
+
predictor = SamPredictor(sam)
|
| 17 |
+
|
| 18 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
| 19 |
+
"""
|
| 20 |
+
data args:
|
| 21 |
+
inputs (:obj: `str` | `PIL.Image` | `np.array`)
|
| 22 |
+
kwargs
|
| 23 |
+
Return:
|
| 24 |
+
A :obj:`list` | `dict`: will be serialized and returned
|
| 25 |
+
"""
|
| 26 |
+
|
| 27 |
+
inputs = data.pop("inputs", data)
|
| 28 |
+
image_url = inputs.pop("imageUrl", none)
|
| 29 |
+
|
| 30 |
+
if not image_url:
|
| 31 |
+
return jsonify({"error": "image_url not provided"}), 400
|
| 32 |
+
|
| 33 |
+
try:
|
| 34 |
+
response = requests.get(image_url)
|
| 35 |
+
response.raise_for_status()
|
| 36 |
+
image = response.content
|
| 37 |
+
except requests.RequestException as e:
|
| 38 |
+
return jsonify({"error": f"Error downloading image: {str(e)}"}), 500
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
predictor.set_image(image)
|
| 42 |
+
|
| 43 |
+
image_embedding = predictor.get_image_embedding().cpu().numpy().toList()
|
| 44 |
+
|
| 45 |
+
return jsonify(image_embedding)
|
| 46 |
+
|
| 47 |
+
# pseudo
|
| 48 |
+
# self.model(input)
|
requirements.txt
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# -f https://download.pytorch.org/whl/cu117/torch_stable.html
|
| 2 |
+
# torch
|
| 3 |
+
# torchvision
|
| 4 |
+
git+https://github.com/ara-vardanyan/segment-anything.git
|
| 5 |
+
flask
|
| 6 |
+
requests
|
test_handler.py
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from handler import EndpointHandler
|
| 2 |
+
|
| 3 |
+
# init handler
|
| 4 |
+
my_handler = EndpointHandler(path=".")
|
| 5 |
+
|
| 6 |
+
# prepare sample payload
|
| 7 |
+
payload = {"inputs": "I am quite excited how this will turn out", "imageUrl": "https://res.cloudinary.com/dvfgdnfzd/image/upload/v1693510414/nvae1t0lvgzavfkgb45j.png"}
|
| 8 |
+
|
| 9 |
+
# test the handler
|
| 10 |
+
payload=my_handler(payload)
|
| 11 |
+
|
| 12 |
+
# show results
|
| 13 |
+
print("payload", payload)
|