feat(machine-learning)!: move machine learning to Python based image (#1774)

BREAKING CHANGES
* Users have to update the docker-compose file, machine-learning portion.
* Temporary dropping machine-learning support for Arm64 and Armv7

.github/workflows/docker.yml

@@ -16,6 +16,7 @@ jobs:
       # Prevent a failure in one image from stopping the other builds
       fail-fast: false
       matrix:
+        platforms: ["linux/arm/v7,linux/amd64,linux/arm64"]
         include:
           - context: "server"
             image: "immich-server"
@@ -23,6 +24,7 @@ jobs:
             image: "immich-web"
           - context: "machine-learning"
             image: "immich-machine-learning"
+            platforms: "linux/amd64"
           - context: "nginx"
             image: "immich-proxy"
 
@@ -92,7 +94,7 @@ jobs:
         uses: docker/build-push-action@v4.0.0
         with:
           context: ${{ matrix.context }}
-          platforms: linux/arm/v7,linux/amd64,linux/arm64
+          platforms: ${{ matrix.platform }}
           # Skip pushing when PR from a fork
           push: ${{ !github.event.pull_request.head.repo.fork }}
           cache-from: type=registry,ref=ghcr.io/${{ github.repository_owner }}/immich-build-cache:${{matrix.image}}

docker/docker-compose.dev.yml

@@ -30,18 +30,20 @@ services:
     build:
       context: ../machine-learning
       dockerfile: Dockerfile
-      target: builder
+    command: python main.py
-    command: npm run start:dev
+    ports:
+      - 3003:3003
     volumes:
-      - ../machine-learning:/usr/src/app
+      - ../machine-learning/src:/usr/src/app
       - ${UPLOAD_LOCATION}:/usr/src/app/upload
-      - /usr/src/app/node_modules
+      - model-cache:/cache
     env_file:
       - .env
     environment:
       - NODE_ENV=development
     depends_on:
       - database
+    restart: always
 
   immich-microservices:
     container_name: immich_microservices
@@ -126,3 +128,4 @@ services:
 
 volumes:
   pgdata:
+  model-cache:

docker/docker-compose.staging.yml

@@ -1,95 +0,0 @@
-version: "3.8"
-
-services:
-  immich-server:
-    container_name: immich_server
-    image: altran1502/immich-server:staging
-    entrypoint: ["/bin/sh", "./start-server.sh"]
-    volumes:
-      - ${UPLOAD_LOCATION}:/usr/src/app/upload
-    env_file:
-      - .env
-    environment:
-      - NODE_ENV=production
-    depends_on:
-      - redis
-      - database
-    restart: always
-
-  immich-microservices:
-    container_name: immich_microservices
-    image: altran1502/immich-server:staging
-    entrypoint: ["/bin/sh", "./start-microservices.sh"]
-    volumes:
-      - ${UPLOAD_LOCATION}:/usr/src/app/upload
-    env_file:
-      - .env
-    environment:
-      - NODE_ENV=production
-    depends_on:
-      - redis
-      - database
-    restart: always
-
-  immich-machine-learning:
-    container_name: immich_machine_learning
-    image: altran1502/immich-machine-learning:staging
-    entrypoint: ["/bin/sh", "./entrypoint.sh"]
-    volumes:
-      - ${UPLOAD_LOCATION}:/usr/src/app/upload
-    env_file:
-      - .env
-    environment:
-      - NODE_ENV=production
-    depends_on:
-      - database
-    restart: always
-
-  immich-web:
-    container_name: immich_web
-    image: altran1502/immich-web:staging
-    entrypoint: ["/bin/sh", "./entrypoint.sh"]
-    env_file:
-      - .env
-    environment:
-      # Rename these values for svelte public interface
-      - PUBLIC_IMMICH_SERVER_URL=${IMMICH_SERVER_URL}
-      - PUBLIC_IMMICH_API_URL_EXTERNAL=${IMMICH_API_URL_EXTERNAL}
-    restart: always
-
-  redis:
-    container_name: immich_redis
-    image: redis:6.2
-    restart: always
-
-  database:
-    container_name: immich_postgres
-    image: postgres:14
-    env_file:
-      - .env
-    environment:
-      POSTGRES_PASSWORD: ${DB_PASSWORD}
-      POSTGRES_USER: ${DB_USERNAME}
-      POSTGRES_DB: ${DB_DATABASE_NAME}
-      PG_DATA: /var/lib/postgresql/data
-    volumes:
-      - pgdata:/var/lib/postgresql/data
-    restart: always
-
-  immich-proxy:
-    container_name: immich_proxy
-    image: altran1502/immich-proxy:staging
-    environment:
-      # Make sure these values get passed through from the env file
-      - IMMICH_SERVER_URL
-      - IMMICH_WEB_URL
-    ports:
-      - 2283:8080
-    logging:
-      driver: none
-    depends_on:
-      - immich-server
-    restart: always
-
-volumes:
-  pgdata:

docker/docker-compose.yml

@@ -4,7 +4,7 @@ services:
   immich-server:
     container_name: immich_server
     image: altran1502/immich-server:release
-    entrypoint: ["/bin/sh", "./start-server.sh"]
+    entrypoint: [ "/bin/sh", "./start-server.sh" ]
     volumes:
       - ${UPLOAD_LOCATION}:/usr/src/app/upload
     env_file:
@@ -19,7 +19,7 @@ services:
   immich-microservices:
     container_name: immich_microservices
     image: altran1502/immich-server:release
-    entrypoint: ["/bin/sh", "./start-microservices.sh"]
+    entrypoint: [ "/bin/sh", "./start-microservices.sh" ]
     volumes:
       - ${UPLOAD_LOCATION}:/usr/src/app/upload
     env_file:
@@ -34,9 +34,10 @@ services:
   immich-machine-learning:
     container_name: immich_machine_learning
     image: altran1502/immich-machine-learning:release
-    entrypoint: ["/bin/sh", "./entrypoint.sh"]
+    command: [ "python", "src/main.py" ]
     volumes:
       - ${UPLOAD_LOCATION}:/usr/src/app/upload
+      - model-cache:/cache
     env_file:
       - .env
     environment:
@@ -48,7 +49,7 @@ services:
   immich-web:
     container_name: immich_web
     image: altran1502/immich-web:release
-    entrypoint: ["/bin/sh", "./entrypoint.sh"]
+    entrypoint: [ "/bin/sh", "./entrypoint.sh" ]
     env_file:
       - .env
     restart: always
@@ -89,3 +90,4 @@ services:
 
 volumes:
   pgdata:
+  model-cache:

machine-learning/.dockerignore

@@ -1,4 +1 @@
-node_modules/
+venv/
-upload/
-dist/
-

machine-learning/.eslintrc.js

@@ -1,24 +0,0 @@
-module.exports = {
-  parser: '@typescript-eslint/parser',
-  parserOptions: {
-    project: 'tsconfig.json',
-    sourceType: 'module',
-  },
-  plugins: ['@typescript-eslint/eslint-plugin'],
-  extends: [
-    'plugin:@typescript-eslint/recommended',
-    'plugin:prettier/recommended',
-  ],
-  root: true,
-  env: {
-    node: true,
-    jest: true,
-  },
-  ignorePatterns: ['.eslintrc.js'],
-  rules: {
-    '@typescript-eslint/interface-name-prefix': 'off',
-    '@typescript-eslint/explicit-function-return-type': 'off',
-    '@typescript-eslint/explicit-module-boundary-types': 'off',
-    '@typescript-eslint/no-explicit-any': 'off',
-  },
-};

machine-learning/.gitignore

@@ -1,37 +1,4 @@
-# compiled output
-/dist
-/node_modules
-
-# Logs
-logs
-*.log
-npm-debug.log*
-pnpm-debug.log*
-yarn-debug.log*
-yarn-error.log*
-lerna-debug.log*
-
-# OS
-.DS_Store
-
-# Tests
-/coverage
-/.nyc_output
-
-# IDEs and editors
-/.idea
-.project
-.classpath
-.c9/
-*.launch
-.settings/
-*.sublime-workspace
-
-# IDE - VSCode
-.vscode/*
-!.vscode/settings.json
-!.vscode/tasks.json
-!.vscode/launch.json
-!.vscode/extensions.json
-
 upload/
+venv/
+__pycache__/
+model-cache/

machine-learning/.prettierrc

@@ -1,4 +0,0 @@
-{
-  "singleQuote": true,
-  "trailingComma": "all"
-}

machine-learning/Dockerfile

@@ -1,34 +1,11 @@
-FROM node:16-bullseye-slim as builder
+FROM python:3.10
 
-ARG DEBIAN_FRONTEND=noninteractive
+ENV TRANSFORMERS_CACHE=/cache
 
 WORKDIR /usr/src/app
 
-RUN apt-get update
+RUN pip install --user --no-cache-dir torch==1.13.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
-RUN apt-get install gcc g++ make cmake python3 python3-pip -y
+RUN pip install --user transformers tqdm numpy scikit-learn scipy nltk sentencepiece flask Pillow
-
+RUN pip install --user --no-deps sentence-transformers
-COPY package.json package-lock.json ./
-
-RUN npm ci
-RUN npm rebuild @tensorflow/tfjs-node --build-from-source
 
 COPY . .
-
-FROM builder as prod
-
-RUN npm run build
-RUN npm prune --omit=dev
-
-FROM node:16-bullseye-slim
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-WORKDIR /usr/src/app
-
-COPY --from=prod /usr/src/app/node_modules ./node_modules
-COPY --from=prod /usr/src/app/dist ./dist
-
-COPY package.json package-lock.json ./
-COPY entrypoint.sh ./
-
-# CMD [ "node", "dist/main" ]

machine-learning/LICENSE

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2022 Hau Tran
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

machine-learning/README.md

@@ -1,4 +1,5 @@
 
-# Microservices for Immich
+# Immich Machine Learning
 
-## Image Classifier
+- Object Detection
+- Image Classification

machine-learning/entrypoint.sh

@@ -1,4 +0,0 @@
-#! /bin/sh
-# npm run typeorm migration:run
-# npm run start:prod
-exec node dist/main.js

machine-learning/nest-cli.json

@@ -1,4 +0,0 @@
-{
-  "collection": "@nestjs/schematics",
-  "sourceRoot": "src"
-}

machine-learning/package.json

@@ -1,71 +0,0 @@
-{
-  "name": "nest_microservices",
-  "version": "0.0.1",
-  "description": "",
-  "author": "",
-  "private": true,
-  "license": "UNLICENSED",
-  "scripts": {
-    "prebuild": "rimraf dist",
-    "build": "nest build",
-    "format": "prettier --write \"src/**/*.ts\" \"test/**/*.ts\"",
-    "start": "nest start",
-    "start:dev": "nest start --watch",
-    "start:debug": "nest start --debug --watch",
-    "start:prod": "node dist/main",
-    "lint": "eslint \"{src,apps,libs,test}/**/*.ts\" --fix",
-    "test": "jest",
-    "test:watch": "jest --watch",
-    "test:cov": "jest --coverage",
-    "test:debug": "node --inspect-brk -r tsconfig-paths/register -r ts-node/register node_modules/.bin/jest --runInBand",
-    "test:e2e": "jest --config ./test/jest-e2e.json"
-  },
-  "dependencies": {
-    "@nestjs/common": "^8.0.0",
-    "@nestjs/core": "^8.0.0",
-    "@tensorflow-models/coco-ssd": "^2.2.2",
-    "@tensorflow-models/mobilenet": "^2.1.0",
-    "@tensorflow/tfjs-node": "^3.19.0"
-  },
-  "devDependencies": {
-    "@nestjs/cli": "^8.2.4",
-    "@nestjs/schematics": "^8.0.0",
-    "@nestjs/testing": "^8.0.0",
-    "@types/express": "^4.17.13",
-    "@types/jest": "27.4.1",
-    "@types/node": "^16.0.0",
-    "@types/supertest": "^2.0.11",
-    "@typescript-eslint/eslint-plugin": "^5.0.0",
-    "@typescript-eslint/parser": "^5.0.0",
-    "eslint": "^8.0.1",
-    "eslint-config-prettier": "^8.3.0",
-    "eslint-plugin-prettier": "^4.0.0",
-    "jest": "^27.2.5",
-    "prettier": "^2.3.2",
-    "rimraf": "^3.0.2",
-    "source-map-support": "^0.5.20",
-    "supertest": "^6.1.3",
-    "ts-jest": "^27.0.3",
-    "ts-loader": "^9.2.3",
-    "ts-node": "^10.0.0",
-    "tsconfig-paths": "^3.10.1",
-    "typescript": "^4.3.5"
-  },
-  "jest": {
-    "moduleFileExtensions": [
-      "js",
-      "json",
-      "ts"
-    ],
-    "rootDir": "src",
-    "testRegex": ".*\\.spec\\.ts$",
-    "transform": {
-      "^.+\\.(t|j)s$": "ts-jest"
-    },
-    "collectCoverageFrom": [
-      "**/*.(t|j)s"
-    ],
-    "coverageDirectory": "../coverage",
-    "testEnvironment": "node"
-  }
-}

machine-learning/src/app.module.ts

@@ -1,10 +0,0 @@
-import { Module } from '@nestjs/common';
-import { ImageClassifierModule } from './image-classifier/image-classifier.module';
-import { ObjectDetectionModule } from './object-detection/object-detection.module';
-
-@Module({
-  imports: [ImageClassifierModule, ObjectDetectionModule],
-  controllers: [],
-  providers: [],
-})
-export class AppModule {}

machine-learning/src/image-classifier/image-classifier.controller.ts

@@ -1,14 +0,0 @@
-import { Body, Controller, Post } from '@nestjs/common';
-import { ImageClassifierService } from './image-classifier.service';
-
-@Controller('image-classifier')
-export class ImageClassifierController {
-  constructor(
-    private readonly imageClassifierService: ImageClassifierService,
-  ) { }
-
-  @Post('/tag-image')
-  async tagImage(@Body('thumbnailPath') thumbnailPath: string) {
-    return await this.imageClassifierService.tagImage(thumbnailPath);
-  }
-}

machine-learning/src/image-classifier/image-classifier.module.ts

@@ -1,9 +0,0 @@
-import { Module } from '@nestjs/common';
-import { ImageClassifierService } from './image-classifier.service';
-import { ImageClassifierController } from './image-classifier.controller';
-
-@Module({
-  controllers: [ImageClassifierController],
-  providers: [ImageClassifierService],
-})
-export class ImageClassifierModule {}

machine-learning/src/image-classifier/image-classifier.service.ts

@@ -1,49 +0,0 @@
-import { Injectable, Logger } from '@nestjs/common';
-import * as mobilenet from '@tensorflow-models/mobilenet';
-import * as cocoSsd from '@tensorflow-models/coco-ssd';
-import * as tf from '@tensorflow/tfjs-node';
-import * as fs from 'fs';
-
-@Injectable()
-export class ImageClassifierService {
-  private readonly MOBILENET_VERSION = 2;
-  private readonly MOBILENET_ALPHA = 1.0;
-
-  private mobileNetModel: mobilenet.MobileNet;
-
-  constructor() {
-    Logger.log(
-      `Running Node TensorFlow Version : ${tf.version['tfjs']}`,
-      'ImageClassifier',
-    );
-    mobilenet
-      .load({
-        version: this.MOBILENET_VERSION,
-        alpha: this.MOBILENET_ALPHA,
-      })
-      .then((mobilenetModel) => (this.mobileNetModel = mobilenetModel));
-  }
-
-  async tagImage(thumbnailPath: string) {
-    try {
-      const isExist = fs.existsSync(thumbnailPath);
-      if (isExist) {
-        const tags = [];
-        const image = fs.readFileSync(thumbnailPath);
-        const decodedImage = tf.node.decodeImage(image, 3) as tf.Tensor3D;
-        const predictions = await this.mobileNetModel.classify(decodedImage);
-
-        for (const prediction of predictions) {
-          if (prediction.probability >= 0.1) {
-            tags.push(...prediction.className.split(',').map((e) => e.trim()));
-          }
-        }
-
-        tf.dispose(decodedImage);
-        return tags;
-      }
-    } catch (e) {
-      console.log('Error reading file ', e);
-    }
-  }
-}

machine-learning/src/main.py

@@ -0,0 +1,61 @@
+import os
+from flask import Flask, request
+from transformers import pipeline
+
+
+server = Flask(__name__)
+
+
+classifier = pipeline(
+    task="image-classification",
+    model="microsoft/resnet-50"
+)
+
+detector = pipeline(
+    task="object-detection",
+    model="hustvl/yolos-tiny"
+)
+
+
+# Environment resolver
+is_dev = os.getenv('NODE_ENV') == 'development'
+server_port = os.getenv('MACHINE_LEARNING_PORT') or 3003
+
+
+@server.route("/ping")
+def ping():
+    return "pong"
+
+
+@server.route("/object-detection/detect-object", methods=['POST'])
+def object_detection():
+    assetPath = request.json['thumbnailPath']
+    return run_engine(detector, assetPath), 201
+
+
+@server.route("/image-classifier/tag-image", methods=['POST'])
+def image_classification():
+    assetPath = request.json['thumbnailPath']
+    return run_engine(classifier, assetPath), 201
+
+
+def run_engine(engine, path):
+    result = []
+    predictions = engine(path)
+
+    for index, pred in enumerate(predictions):
+        tags = pred['label'].split(', ')
+        if (index == 0):
+            result = tags
+        else:
+            if (pred['score'] > 0.5):
+                result = [*result, *tags]
+
+    if (len(result) > 1):
+        result = list(set(result))
+
+    return result
+
+
+if __name__ == "__main__":
+    server.run(debug=is_dev, host='0.0.0.0', port=server_port)

machine-learning/src/main.ts

@@ -1,27 +0,0 @@
-import { NestFactory } from '@nestjs/core';
-import { AppModule } from './app.module';
-import { Logger } from '@nestjs/common';
-
-async function bootstrap() {
-  const app = await NestFactory.create(AppModule);
-
-  const port = Number(process.env.MACHINE_LEARNING_PORT) || 3003;
-
-  await app.listen(port, () => {
-    if (process.env.NODE_ENV == 'development') {
-      Logger.log(
-        'Running Immich Machine Learning in DEVELOPMENT environment',
-        'IMMICH MICROSERVICES',
-      );
-    }
-
-    if (process.env.NODE_ENV == 'production') {
-      Logger.log(
-        'Running Immich Machine Learning in PRODUCTION environment',
-        'IMMICH MICROSERVICES',
-      );
-    }
-  });
-}
-
-bootstrap();

machine-learning/src/object-detection/object-detection.controller.ts

@@ -1,15 +0,0 @@
-import { Body, Controller, Post } from '@nestjs/common';
-import { ObjectDetectionService } from './object-detection.service';
-import { Logger } from '@nestjs/common';
-
-@Controller('object-detection')
-export class ObjectDetectionController {
-  constructor(
-    private readonly objectDetectionService: ObjectDetectionService,
-  ) { }
-
-  @Post('/detect-object')
-  async detectObject(@Body('thumbnailPath') thumbnailPath: string) {
-    return await this.objectDetectionService.detectObject(thumbnailPath);
-  }
-}

machine-learning/src/object-detection/object-detection.module.ts

@@ -1,9 +0,0 @@
-import { Module } from '@nestjs/common';
-import { ObjectDetectionService } from './object-detection.service';
-import { ObjectDetectionController } from './object-detection.controller';
-
-@Module({
-  controllers: [ObjectDetectionController],
-  providers: [ObjectDetectionService],
-})
-export class ObjectDetectionModule {}

machine-learning/src/object-detection/object-detection.service.ts

@@ -1,39 +0,0 @@
-import { Injectable, Logger } from '@nestjs/common';
-import * as cocoSsd from '@tensorflow-models/coco-ssd';
-import * as tf from '@tensorflow/tfjs-node';
-import * as fs from 'fs';
-
-@Injectable()
-export class ObjectDetectionService {
-  private cocoSsdModel: cocoSsd.ObjectDetection;
-
-  constructor() {
-    Logger.log(
-      `Running Node TensorFlow Version : ${tf.version['tfjs']}`,
-      'ObjectDetection',
-    );
-    cocoSsd.load().then((model) => (this.cocoSsdModel = model));
-  }
-  async detectObject(thumbnailPath: string) {
-    try {
-      const isExist = fs.existsSync(thumbnailPath);
-      if (isExist) {
-        const tags = new Set();
-        const image = fs.readFileSync(thumbnailPath);
-        const decodedImage = tf.node.decodeImage(image, 3) as tf.Tensor3D;
-        const predictions = await this.cocoSsdModel.detect(decodedImage);
-
-        for (const result of predictions) {
-          if (result.score > 0.5) {
-            tags.add(result.class);
-          }
-        }
-
-        tf.dispose(decodedImage);
-        return [...tags];
-      }
-    } catch (e) {
-      console.log('Error reading file ', e);
-    }
-  }
-}

machine-learning/tsconfig.build.json

@@ -1,4 +0,0 @@
-{
-  "extends": "./tsconfig.json",
-  "exclude": ["node_modules", "test", "dist", "**/*spec.ts"]
-}

machine-learning/tsconfig.json

@@ -1,21 +0,0 @@
-{
-  "compilerOptions": {
-    "module": "commonjs",
-    "declaration": true,
-    "removeComments": true,
-    "emitDecoratorMetadata": true,
-    "experimentalDecorators": true,
-    "allowSyntheticDefaultImports": true,
-    "target": "es2017",
-    "sourceMap": true,
-    "outDir": "./dist",
-    "baseUrl": "./",
-    "incremental": true,
-    "skipLibCheck": true,
-    "strictNullChecks": false,
-    "noImplicitAny": false,
-    "strictBindCallApply": false,
-    "forceConsistentCasingInFileNames": false,
-    "noFallthroughCasesInSwitch": false
-  }
-}

server/apps/microservices/src/processors/machine-learning.processor.ts

@@ -33,7 +33,6 @@ export class MachineLearningProcessor {
       const smartInfo = new SmartInfoEntity();
       smartInfo.assetId = asset.id;
       smartInfo.tags = [...res.data];
-
       await this.smartInfoRepository.upsert(smartInfo, {
         conflictPaths: ['assetId'],
       });