Merge pull request #48 from datalab-mi/develop

v1.1 Develop

.github/release-drafter.yml

@@ -5,6 +5,9 @@ categories:
   - title: '🐛 Bug Fixes'
     labels:
       - 'bug'
+  - title: '🔥 Enhancements'
+    labels:
+      - 'enhancement'
 autolabeler:
   - label: 'bug'
     branch:
@@ -15,6 +18,9 @@ autolabeler:
     branch:
       - '/feature\/.+/'
       - '/feat\/.+/'
+  - label: 'enhancement'
+    branch:
+      - '/enhancement\/.+/'
 change-template: '- $TITLE @$AUTHOR (#$NUMBER)'
 change-title-escapes: '\<*_&' # You can add # and @ to disable mentions, and add ` to disable code blocks.
 template: |

.github/workflows/pr.yml

@@ -0,0 +1,43 @@
+name: CI for pr
+on:
+  pull_request:
+    types: [opened, reopened, synchronize]
+
+jobs:
+  tag-pr:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Check out the repo
+      uses: actions/checkout@v2
+    - name: Tag PR
+      uses: release-drafter/release-drafter@v5
+      with:
+        name: PR ${{ github.ref }}
+        prerelease: true
+      env:
+        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  check-files-modified:
+    runs-on: ubuntu-latest
+    outputs:
+      changed_files: ${{ steps.changed-files.outputs.all_modified_files }}
+    steps:
+    - uses: actions/checkout@v2
+    - name: Verify Changed files
+      uses: tj-actions/[email protected]
+      id: changed-files
+      with:
+        base_sha: ${{ github.event.pull_request.base.sha }}
+
+  build-backend-and-test:
+    runs-on: ubuntu-latest
+    needs: check-files-modified
+    if: ( contains(needs.check-files-modified.outputs.changed_files, 'backend') )
+    steps:
+    - uses: actions/checkout@v2
+    - name: Build docker image for tests
+      run: docker build --target test -t basegun-back:tests backend/
+    - name: Start container
+      run: docker run --rm --name basegun_back_test -d basegun-back:tests
+    - name: Run tests
+      run: docker exec basegun_back_test python -m unittest discover -v

.github/workflows/publish-image-github.yml → .github/workflows/release.yml

@@ -1,16 +1,14 @@
-name: Push Docker image to GitHub container
+name: CI for release
 on:
   push:
     tags:
       - 'v*'
-  pull_request:
-    types: [opened, reopened, synchronize]
+
 jobs:
   release_github:
     runs-on: ubuntu-latest
     steps:
-    - name: Check out the repo
-      uses: actions/checkout@v2
+    - uses: actions/checkout@v2
     - name: Get the version
       id: get_version
       run: echo ::set-output name=VERSION::${GITHUB_REF/refs\/tags\//}
@@ -24,10 +22,8 @@ jobs:
 
   build_and_push:
     runs-on: ubuntu-latest
-    if: startsWith(github.ref, 'refs/tags/v')
     steps:
-    - name: Check out the repo
-      uses: actions/checkout@v2
+    - uses: actions/checkout@v2
     - name: Build the Docker image
       run: TAG=action docker-compose -f docker-compose-prod.yml build
     - name: Login to GHCR

.gitignore

@@ -6,4 +6,4 @@ env/
 __pycache__
 weights/
 temp/
-openrc.sh
+openrc.sh

Makefile

@@ -1,7 +1,7 @@
 SHELL	:= /bin/bash
 DOCKER	:= $(shell type -p docker)
 DC		:= $(shell type -p docker-compose)
-TAG		:= 1.0
+TAG		:= 1.1
 
 export
 
@@ -40,4 +40,4 @@ tag: show-current-tag
 
 untag: show-current-tag
 	git tag -d v${TAG}
-	git push --delete v${TAG}
+	git push --delete origin v${TAG}

backend/Dockerfile

@@ -1,10 +1,11 @@
-FROM python:3.9-slim-buster
+FROM python:3.9-slim-buster as base
 WORKDIR /app
 
 # librairies necessary for image processing
 RUN apt update && apt install -y \
     libgl1-mesa-glx \
     libglib2.0-0 \
+    curl \
     && rm -rf /var/lib/apt/lists/*
 
 # install python libraries
@@ -14,5 +15,17 @@ RUN pip install --upgrade pip \
     && rm -r /root/.cache
 
 # launch website
+RUN curl -o model.pth https://storage.gra.cloud.ovh.net/v1/AUTH_df731a99a3264215b973b3dee70a57af/basegun-public/models/EffB7_2022-03-30_08/EffB7_2022-03-30_08.pth
 COPY src/ src/
-CMD ["uvicorn", "src.main:app", "--reload", "--host", "0.0.0.0", "--port", "5000"]
+RUN mkdir -p src/weights && mv model.pth src/weights/model.pth
+
+FROM base as dev
+CMD ["uvicorn", "src.main:app", "--reload", "--host", "0.0.0.0", "--port", "5000"]
+
+FROM base as test
+RUN pip install requests && rm -r /root/.cache
+COPY tests/ tests/
+CMD ["uvicorn", "src.main:app", "--reload", "--host", "0.0.0.0", "--port", "5000"]
+
+FROM base as prod
+CMD ["uvicorn", "src.main:app", "--host", "0.0.0.0", "--port", "5000"]

backend/README.md

@@ -4,18 +4,17 @@ From this folder
 
 ## Install
 
-**Download model** EfficientNetB7 from OVH Cloud, put it in src/weights/ :
-```
-ovhai data download GRA basegun-public models/EffB7_2022-03-01_17/EffB7_2022-03-01_17.pth -o src/weights/
-```
-
 ### Without Docker
 1. Activate python environment with Python=3.8 (conda or pyenv)
-2. `pip install -r requirements.txt -f https://download.pytorch.org/whl/cpu/torch_stable.html`
+2. Download model `curl https://storage.gra.cloud.ovh.net/v1/AUTH_df731a99a3264215b973b3dee70a57af/basegun-public/models/EffB7_2022-03-01_17/EffB7_2022-03-01_17.pth -o src/weights/model.pth`
+3. `pip install -r requirements.txt -f https://download.pytorch.org/whl/cpu/torch_stable.html`
 
 ### With Docker
-1. Create variable HTTP_PROXY in your shell for the Ministry proxy
-2. Build image `docker build -t basegun-back:dev .`
+Build image
+```bash
+docker build --target dev -t basegun-back:dev .
+```
+If you are in a network blocked with proxy, remember to add arg `--build_arg https_proxy` where `https_proxy` is a variable already set in your env.
 
 
 ## Run
@@ -31,6 +30,11 @@ uvicorn src.main:app --reload --host 0.0.0.0 --port 5000
 
 ### With Docker
 ```bash
-docker run --rm -d -p 5000:5000 basegun-back:dev -e PATH_IMGS=/tmp/basegun/
+docker run --rm -d -p 5000:5000 --name basegun_back basegun-back:dev -e PATH_IMGS=/tmp/basegun/
 ```
+Remember afterwards to stop container `docker stop basegun_back`
 
+## Run tests
+1. Build image to target test `docker build --target test -t basegun-back:test .`
+2. Start container `docker run --rm --name basegun_back_test -d basegun-back:test`
+3. Execute tests `docker exec basegun_back_test python -m unittest discover -v`

backend/src/main.py

@@ -1,8 +1,11 @@
-import shutil, os
+import shutil
+import os
+from uuid import uuid4
 from fastapi import FastAPI, File, UploadFile, HTTPException
+from fastapi.responses import PlainTextResponse, FileResponse
 from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import FileResponse
-from src.model import load_model_inference, test_image
+from PIL import UnidentifiedImageError
+from src.model import load_model_inference, predict_image
 
 app = FastAPI()
 
@@ -16,7 +19,7 @@
 
 MODEL_PATH = os.path.join(
             os.path.dirname(os.path.abspath(__file__)),
-            "weights/EffB7_2022-03-01_17.pth")
+            "weights/model.pth")
 
 if "PATH_IMGS" in os.environ:
     PATH_IMGS = os.environ["PATH_IMGS"]
@@ -25,6 +28,7 @@
             os.path.dirname(os.path.abspath(__file__)),
             "../../frontend/public/temp"))
     print("WARNING: The variable PATH_IMGS is not set. Using", PATH_IMGS)
+os.makedirs(PATH_IMGS, exist_ok = True)
 
 # allow requests from front-end
 app.add_middleware(
@@ -40,19 +44,29 @@
     model = load_model_inference(MODEL_PATH)
 
 
-@app.get("/")
+@app.get("/", response_class=PlainTextResponse)
 def home():
     return "Basegun backend"
 
 
 @app.post("/upload")
 async def imageupload(image: UploadFile = File(...)):
     if model:
-        input_path = os.path.join(PATH_IMGS, image.filename)
+        input_path = os.path.join(
+                        PATH_IMGS, # store image in PATH_IMGS folder
+                        # rename with uuid for secure filename but keep original file ext
+                        str(uuid4()) + os.path.splitext(image.filename)[1]
+                    )
         with open(f'{input_path}', "wb") as buffer:
             shutil.copyfileobj(image.file, buffer)
-        label, confidence = test_image(model, input_path)
-        print("Finished processing, result:", input_path, label, confidence)
+        try:
+            label, confidence = predict_image(model, input_path)
+            print("Finished processing, result:", input_path, label, confidence)
+        except UnidentifiedImageError:
+            raise HTTPException(status_code=400, detail="Corrupted image file")
+        except Exception as e:
+            raise HTTPException(status_code=500, detail=str(e))
+
     else:
         raise HTTPException(status_code=404, detail="Model not found")
     return {"file_name": input_path, "label": label, "confidence": confidence}

backend/src/model.py

@@ -1,6 +1,6 @@
 import os
+from typing import Union
 from PIL import Image
-from torch import Tensor
 import numpy as np
 import torch
 import torchvision.models as Model
@@ -83,18 +83,33 @@ def __call__(self, image):
 
 
 def build_model(model: Model) -> Model:
-    # freeze first layers
+    """Create the model structure
+
+    Args:
+        model (Model): raw torchvision model
+
+    Returns:
+        Model: modified model with classification layer size len(CLASSES)
+    """
+    # freeze all layers except classification - very important
     for param in model.parameters():
         param.requires_grad = False
-    # Parameters of newly constructed modules have requires_grad=True by default
+    # replace last layer of model for our number of classes
     num_ftrs = model.classifier[1].in_features
-    # to try later : add batch normalization and dropout
     model.classifier[1] = torch.nn.Linear(num_ftrs, len(CLASSES))
     model = model.to(device)
     return model
 
 
 def load_model_inference(state_dict_path: str) -> Model:
+    """Load model structure and weights
+
+    Args:
+        state_dict_path (str): path to model (.pth file)
+
+    Returns:
+        Model: loaded model ready for prediction
+    """
     model = build_model(MODEL_TORCH())
     # Initialize model with the pretrained weights
     model.load_state_dict(torch.load(state_dict_path, map_location=device)['model_state_dict'])
@@ -104,19 +119,39 @@ def load_model_inference(state_dict_path: str) -> Model:
     return model
 
 
-loader =  transforms.Compose([
-            ConvertRgb(),
-            Rescale(INPUT_SIZE),
-            RandomPad(INPUT_SIZE),
-            transforms.ToTensor(),
-            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
-        ])
+def prepare_input(image: Image) -> torch.Tensor:
+    """Convert a PIL Image to model-compatible input
+
+    Args:
+        image (Image): input image
+
+    Returns:
+        torch.Tensor: converted image
+        (shape (1, 3, size, size), normalized on ImageNet)
+    """
+    loader = transforms.Compose([
+        ConvertRgb(),
+        Rescale(INPUT_SIZE),
+        RandomPad(INPUT_SIZE),
+        transforms.ToTensor(),
+        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+    ])
+    image = loader(image).float()
+    return image.unsqueeze(0).to(device)
 
 
-def test_image(model, path):
+def predict_image(model: Model, path: str) -> Union[str, float]:
+    """Run the model prediction on an image
+
+    Args:
+        model (Model): classification model
+        path (str): path to input image
+
+    Returns:
+        Union[str, float]: (label, confidence) of best class predicted
+    """
     im = Image.open(path)
-    image = loader(im).float()
-    image = image.unsqueeze(0).to(device)
+    image = prepare_input(im)
     output = model(image)
     probs = torch.nn.functional.softmax(output, dim=1).detach().numpy()[0]
     res = [(CLASSES[i], round(probs[i]*100,2)) for i in range(len(CLASSES))]

backend/tests/test_api.py

@@ -0,0 +1,42 @@
+import sys
+import unittest
+import os
+import requests
+from PIL import Image, ImageChops
+from src.main import PATH_IMGS
+
+class TestModel(unittest.TestCase):
+    def __init__(self, *args, **kwargs):
+        super(TestModel, self).__init__(*args, **kwargs)
+        self.url = "http://localhost:5000"
+        self.assertTrue(os.path.exists(PATH_IMGS))
+
+    def test_home(self):
+        """Checks that the route / is alive"""
+        r = requests.get(self.url)
+        self.assertEqual(r.text, "Basegun backend")
+
+    def test_upload(self):
+        """Checks that the file upload works properly"""
+        path = os.path.join(
+            os.path.dirname(os.path.abspath(__file__)),
+            "revolver.jpg")
+        with open(path, 'rb') as f:
+            r = requests.post(self.url + "/upload", files={"image": f})
+        res = r.json()
+
+        # checks that the input file has been written
+        self.assertTrue("file_name" in res.keys())
+        output_path = os.path.join(PATH_IMGS, res["file_name"])
+        self.assertTrue(os.path.exists(output_path))
+        # checks that written file is exactly the same as input file
+        with Image.open(path) as image_one:
+            with Image.open(output_path) as image_two:
+                self.assertEqual(image_one.size, image_two.size)
+                diff = ImageChops.difference(image_one, image_two)
+                self.assertFalse(diff.getbbox())
+        # checks that the json result is as expected
+        self.assertTrue("label" in res.keys())
+        self.assertEqual(res["label"], "revolver")
+        self.assertTrue("confidence" in res.keys())
+        self.assertAlmostEqual(res["confidence"], 99.05, places=1)