First commit

Dockerfile

@@ -0,0 +1,22 @@
+FROM python:3.6-slim-stretch
+
+RUN apt update
+RUN apt install -y python3-dev gcc
+
+# Install pytorch and fastai
+RUN pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
+RUN pip install fastai
+
+# Install starlette and uvicorn
+RUN pip install starlette uvicorn python-multipart aiohttp
+
+ADD imgapp.py imgapp.py
+ADD resnet50_EAP_version6.pkl resnet50_EAP_version6.pkl  
+
+# Run it once to trigger resnet download
+RUN python imgapp.py
+
+EXPOSE 8008
+
+# Start the server
+CMD ["python", "imgapp.py", "serve"]

README.md

@@ -0,0 +1,11 @@
+# Image orientation detection
+*App inspired by the work of [Deepanshu Thakur](https://github.com/Deepanshu2017/AlligatorOrCrocodile) and [Simon Willison](https://github.com/simonw/cougar-or-not)*
+
+This webapp was created on behalf of the British Library [Endangered Archive Programme](https://eap.bl.uk/) to determine the orientation of images in their collection. The deep learning model is based on a 50-layer version of [resnet](https://arxiv.org/abs/1512.03385) and was trained on around 650 unique images from the EAP archives using the [fast.ai](https://docs.fast.ai/) library. Each image is duplicated into four representations, consisting of the application of 0, 90, 180 or 270 degrees of anti-clockwise rotation.
+
+The 99MB pre-trained model is included in this package and is found within `resnet50_EAP_version6.pkl`
+
+`imgapp.py` is a [Starlette](https://www.starlette.io/) API server which accepts image uploads or image URLs and runs them against the pre-calculated model.
+
+A `Dockerfile` is included for convenience.
+

imgapp.py

@@ -0,0 +1,99 @@
+from starlette.applications import Starlette
+from starlette.responses import JSONResponse, HTMLResponse, RedirectResponse
+from fastai import *
+from fastai.vision import * 
+
+import torch
+from pathlib import Path
+from io import BytesIO
+import sys
+import uvicorn
+import aiohttp
+import asyncio
+
+
+async def get_bytes(url):
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as response:
+            return await response.read()
+
+
+app = Starlette()
+path = Path(".")
+preTrainedModel="resnet50_EAP_version6.pkl"
+classes = ["0","90","180","270"]
+
+
+
+
+
+data = ImageDataBunch.single_from_classes(
+    path,
+    classes,
+    ds_tfms=get_transforms(),
+    size=224,
+).normalize(imagenet_stats)
+
+learn=load_learner(path,preTrainedModel)
+
+@app.route("/upload", methods=["POST"])
+async def upload(request):
+    data = await request.form()
+    bytes = await (data["file"].read())
+    return predict_image_from_bytes(bytes)
+
+
+@app.route("/classify-url", methods=["GET"])
+async def classify_url(request):
+    bytes = await get_bytes(request.query_params["url"])
+    return predict_image_from_bytes(bytes)
+
+
+def predict_image_from_bytes(bytes):
+    """
+    Predict function called by either classify_url or by upload
+    Returns True class as well as the scores
+    Scores are *not* probabilities - use activation function (softmax, sigmoid, etc)
+    to convert to probabilities!
+
+    """
+
+    img = open_image(BytesIO(bytes))
+    class_, predictions, losses = learn.predict(img) 
+    return JSONResponse({
+        "class": class_,
+        "scores": sorted(
+            zip(learn.data.classes, map(float, losses)),
+            key=lambda p: p[1],
+            reverse=True
+        )
+    })
+
+
+@app.route("/")
+def form(request):
+    return HTMLResponse(
+        """
+        <h3> Image orientation prediction app <h3>
+        <h4> Trained on approximately 650 images, each with four representations<h4>
+        <form action="/upload" method="post" enctype="multipart/form-data">
+            Select image to upload:
+            <input type="file" name="file">
+            <input type="submit" value="Upload Image">
+        </form>
+        Or submit a URL:
+        <form action="/classify-url" method="get">
+            <input type="url" name="url">
+            <input type="submit" value="Fetch and analyze image">
+        </form>
+    """)
+
+
+@app.route("/form")
+def redirect_to_homepage(request):
+    return RedirectResponse("/")
+
+
+if __name__ == "__main__":
+    if "serve" in sys.argv:
+        uvicorn.run(app, host="0.0.0.0", port=8080)