Add authenticated AnyLearning remote inference

.github/workflows/remote-real-model-e2e.yml

@@ -0,0 +1,108 @@
+name: Remote real model E2E
+
+on:
+  pull_request:
+    branches: [develop]
+    paths:
+      - "anylabeling/services/auto_labeling/remote_client.py"
+      - "anylabeling/services/auto_labeling/remote_model.py"
+      - "anylabeling/services/auto_labeling/model_manager.py"
+      - "scripts/validate_remote_inference.py"
+      - ".github/workflows/remote-real-model-e2e.yml"
+  push:
+    branches: [develop]
+    paths:
+      - "anylabeling/services/auto_labeling/remote_client.py"
+      - "anylabeling/services/auto_labeling/remote_model.py"
+      - "anylabeling/services/auto_labeling/model_manager.py"
+      - "scripts/validate_remote_inference.py"
+      - ".github/workflows/remote-real-model-e2e.yml"
+  schedule:
+    - cron: "41 3 * * 1"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: remote-real-model-e2e-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  yolox-s:
+    name: Authenticated YOLOX-S / ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 30
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest, macos-latest]
+    env:
+      ASSET_DIR: _anylearning/tests/fixtures/inference/real_models
+      MODEL_SHA256: c5c2d13e59ae883e6af3b45daea64af4833a4951c92d116ec270d9ddbe998063
+      IMAGE_SHA256: 5a9522051c3cec2bbd2f6323fccba32e8fbf3ddcc2b3e2fd46b04c720bc6f866
+      OPENBLAS_NUM_THREADS: "1"
+      OMP_NUM_THREADS: "1"
+      QT_QPA_PLATFORM: offscreen
+    steps:
+      - uses: actions/checkout@v7
+
+      - name: Check out the pinned AnyLearning server contract
+        uses: actions/checkout@v7
+        with:
+          repository: nrl-ai/anylearning-oss
+          ref: 7cabc1e8caaec070410fbc47e8ee250ae50454ce
+          path: _anylearning
+
+      - uses: actions/setup-python@v7
+        with:
+          python-version: "3.13"
+          cache: pip
+
+      - name: Install Qt system libraries (Linux)
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends \
+            libegl1 libxkbcommon-x11-0 libdbus-1-3 libxcb-cursor0 \
+            libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-randr0 \
+            libxcb-render-util0 libxcb-shape0 libxcb-xinerama0 libxcb-xkb1
+
+      - name: Install PyQt6 (macOS)
+        if: runner.os == 'macOS'
+        run: python -m pip install "PyQt6>=6.7.0"
+
+      - name: Install client and isolated server dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install .
+          python -m pip install "fastapi>=0.141,<1" "pydantic>=2.13,<3" "argon2-cffi>=25.1,<26" "psutil>=7.2,<8" "uvicorn>=0.41,<1"
+
+      - name: Download immutable Apache-2.0 test assets
+        shell: bash
+        run: |
+          set -euo pipefail
+          python _anylearning/scripts/download_verified_file.py \
+            "https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_s.onnx" \
+            "$ASSET_DIR/yolox_s.onnx" --sha256 "$MODEL_SHA256" --max-bytes 104857600
+          python _anylearning/scripts/download_verified_file.py \
+            "https://raw.githubusercontent.com/Megvii-BaseDetection/YOLOX/6ddff4824372906469a7fae2dc3206c7aa4bbaee/assets/dog.jpg" \
+            "$ASSET_DIR/dog.jpg" --sha256 "$IMAGE_SHA256" --max-bytes 10485760
+
+      - name: Run AnyLabeling through the real authenticated server
+        run: >-
+          python scripts/validate_remote_inference.py
+          --anylearning-root _anylearning
+          --manifest _anylearning/tests/fixtures/inference/real_models/yolox_s_official.json
+          --model _anylearning/tests/fixtures/inference/real_models/yolox_s.onnx
+          --image _anylearning/tests/fixtures/inference/real_models/dog.jpg
+          --output-root validation-results/${{ runner.os }}
+
+      - name: Retain visual, result, and timing evidence
+        if: always()
+        uses: actions/upload-artifact@v6
+        with:
+          name: anylabeling-remote-yolox-s-${{ runner.os }}
+          path: validation-results/
+          if-no-files-found: error
+          retention-days: 30

README.md

@@ -42,6 +42,7 @@
   - **SAM 3** (ViT-H) — open-vocabulary segmentation with text prompts
 - [x] Text detection, recognition and KIE (Key Information Extraction) labeling.
 - [x] Hardware acceleration with CUDA, CoreML, DirectML, OpenVINO, and vendor NPU providers.
+- [x] Authenticated shared ONNX inference through an AnyLearning server.
 - [x] Multiple languages available: English, Vietnamese, Chinese.
 
 ### Supported Models
@@ -57,6 +58,8 @@
 
 Required model weights are downloaded automatically on first use.
 
+For centrally hosted models, see the [authenticated remote inference guide](docs/remote_inference.md).
+
 ## Latest Release
 
 [AnyLabeling v0.4.43](https://github.com/vietanhdev/anylabeling/releases/tag/v0.4.43) is the current stable release. It fixes recovery after model download/load failures, skips invalid images during SAM preload, persists grouped shapes with undo support, and closes label files reliably after saving and loading.

anylabeling/services/auto_labeling/__init__.py

@@ -1,4 +1,5 @@
 # Import models to ensure they register themselves via @ModelRegistry.register
+from . import remote_model as remote_model  # noqa: F401
 from . import segment_anything as segment_anything  # noqa: F401
 from . import yolov5 as yolov5  # noqa: F401
 from . import yolov8 as yolov8  # noqa: F401

anylabeling/services/auto_labeling/model.py

@@ -1,7 +1,6 @@
 import logging
 import os
 import socket
-import ssl
 from abc import abstractmethod
 
 import yaml
@@ -18,11 +17,6 @@
 socket.setdefaulttimeout(240)  # Prevent timeout when downloading models
 
 
-ssl._create_default_https_context = (
-    ssl._create_unverified_context
-)  # Prevent issue when downloading models behind a proxy
-
-
 class Model(QObject):
     BASE_DOWNLOAD_URL = "https://github.com/vietanhdev/anylabeling-assets/raw/main/"
 

anylabeling/services/auto_labeling/model_manager.py

@@ -4,7 +4,6 @@
 import os
 import pathlib
 import shutil
-import ssl
 import tempfile
 import time
 import urllib.request
@@ -22,10 +21,6 @@
 
 from .registry import ModelRegistry
 
-ssl._create_default_https_context = (
-    ssl._create_unverified_context
-)  # Prevent issue when downloading models behind a proxy
-
 
 class ModelManager(QObject):
     """Model manager"""
@@ -162,7 +157,11 @@ def on_model_download_finished(self):
             self.model_loaded.emit(self.loaded_model_config)
             self.output_modes_changed.emit(
                 self.loaded_model_config["model"].Meta.output_modes,
-                self.loaded_model_config["model"].Meta.default_output_mode,
+                getattr(
+                    self.loaded_model_config["model"],
+                    "output_mode",
+                    self.loaded_model_config["model"].Meta.default_output_mode,
+                ),
             )
         else:
             self.model_loaded.emit({})
@@ -208,7 +207,8 @@ def load_custom_model(self, config_file):
             "type" not in model_config
             or "display_name" not in model_config
             or "name" not in model_config
-            or model_config["type"] not in ["segment_anything", "yolov5", "yolov8"]
+            or model_config["type"]
+            not in ["remote", "segment_anything", "yolov5", "yolov8"]
         ):
             self._report_model_load_error(
                 self.tr("Error in loading custom model: Invalid config file format.")
@@ -439,7 +439,9 @@ def _load_model(self, model_id):
 
             # Specific logic for interactive models (like SAM) vs detection models
             # Ideally this should be a property of the model class (capabilities)
-            if model_type == "segment_anything":
+            if model_type == "segment_anything" or getattr(
+                model_config["model"], "supports_interactive_prompts", False
+            ):
                 self.auto_segmentation_model_selected.emit()
                 # Request next files for prediction
                 self.request_next_files_requested.emit()
@@ -460,9 +462,13 @@ def set_auto_labeling_marks(self, marks):
         """Set auto labeling marks
         (For example, for segment_anything model, it is the marks for)
         """
-        if (
-            self.loaded_model_config is None
-            or self.loaded_model_config["type"] != "segment_anything"
+        if self.loaded_model_config is None or not (
+            self.loaded_model_config["type"] == "segment_anything"
+            or getattr(
+                self.loaded_model_config["model"],
+                "supports_interactive_prompts",
+                False,
+            )
         ):
             return
         self.loaded_model_config["model"].set_auto_labeling_marks(marks)
@@ -534,8 +540,11 @@ def predict_shapes_threading(self, image, filename=None):
                 self.model_execution_thread is not None
                 and self.model_execution_thread.isRunning()
             ):
-                if hasattr(self.loaded_model_config["model"], "unload"):
-                    self.loaded_model_config["model"].unload()
+                model = self.loaded_model_config["model"]
+                if hasattr(model, "cancel_prediction"):
+                    model.cancel_prediction()
+                elif hasattr(model, "unload"):
+                    model.unload()
 
                 # Wait for the thread to finish
                 self.model_execution_thread.quit()
@@ -573,7 +582,14 @@ def on_next_files_changed(self, next_files):
             return
 
         # Currently only segment_anything model supports this feature
-        if self.loaded_model_config["type"] != "segment_anything":
+        if not (
+            self.loaded_model_config["type"] == "segment_anything"
+            or getattr(
+                self.loaded_model_config["model"],
+                "supports_interactive_prompts",
+                False,
+            )
+        ):
             return
 
         self.loaded_model_config["model"].on_next_files_changed(next_files)