Implementing uint16 reading with tifffile

config/app_config.yaml

@@ -1,57 +0,0 @@
-database:
-  path: data/detections.db
-image_repository:
-  base_path: ''
-  allowed_extensions:
-  - .jpg
-  - .jpeg
-  - .png
-  - .tif
-  - .tiff
-  - .bmp
-models:
-  default_base_model: yolov8s-seg.pt
-  models_directory: data/models
-  base_model_choices:
-  - yolov8s-seg.pt
-  - yolo11s-seg.pt
-training:
-  default_epochs: 100
-  default_batch_size: 16
-  default_imgsz: 1024
-  default_patience: 50
-  default_lr0: 0.01
-  two_stage:
-    enabled: false
-    stage1:
-      epochs: 20
-      lr0: 0.0005
-      patience: 10
-      freeze: 10
-    stage2:
-      epochs: 150
-      lr0: 0.0003
-      patience: 30
-  last_dataset_yaml: /home/martin/code/object_detection/data/datasets/data.yaml
-  last_dataset_dir: /home/martin/code/object_detection/data/datasets
-detection:
-  default_confidence: 0.25
-  default_iou: 0.45
-  max_batch_size: 100
-visualization:
-  bbox_colors:
-    organelle: '#FF6B6B'
-    membrane_branch: '#4ECDC4'
-    default: '#00FF00'
-  bbox_thickness: 2
-  font_size: 12
-export:
-  formats:
-  - csv
-  - json
-  - excel
-  default_format: csv
-logging:
-  level: INFO
-  file: logs/app.log
-  format: '%(asctime)s - %(name)s - %(levelname)s - %(message)s'

src/gui/tabs/training_tab.py

@@ -1303,6 +1303,14 @@ class TrainingTab(QWidget):
         sample_image = self._find_first_image(images_dir)
         if not sample_image:
             return False
+
+        # Do not force an RGB cache for TIFF datasets.
+        # We handle grayscale/16-bit TIFFs via runtime Ultralytics patches that:
+        # - load TIFFs with `tifffile`
+        # - replicate grayscale to 3 channels without quantization
+        # - normalize uint16 correctly during training
+        if sample_image.suffix.lower() in {".tif", ".tiff"}:
+            return False
         try:
             img = Image(sample_image)
             return img.pil_image.mode.upper() != "RGB"

src/model/yolo_wrapper.py

@@ -1,9 +1,13 @@
-"""
-YOLO model wrapper for the microscopy object detection application.
-Provides a clean interface to YOLOv8 for training, validation, and inference.
+"""YOLO model wrapper for the microscopy object detection application.
+
+Notes on 16-bit TIFF support:
+- Ultralytics training defaults assume 8-bit images and normalize by dividing by 255.
+- This project can patch Ultralytics at runtime to decode TIFFs via `tifffile` and
+  normalize `uint16` correctly.
+
+See [`apply_ultralytics_16bit_tiff_patches()`](src/utils/ultralytics_16bit_patch.py:1).
 """
 
-from ultralytics import YOLO
 from pathlib import Path
 from typing import Optional, List, Dict, Callable, Any
 import torch
@@ -11,6 +15,7 @@ import tempfile
 import os
 from src.utils.image import Image
 from src.utils.logger import get_logger
+from src.utils.ultralytics_16bit_patch import apply_ultralytics_16bit_tiff_patches
 
 
 logger = get_logger(__name__)
@@ -31,6 +36,9 @@ class YOLOWrapper:
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
         logger.info(f"YOLOWrapper initialized with device: {self.device}")
 
+        # Apply Ultralytics runtime patches early (before first import/instantiation of YOLO datasets/trainers).
+        apply_ultralytics_16bit_tiff_patches()
+
     def load_model(self) -> bool:
         """
         Load YOLO model from path.
@@ -40,6 +48,9 @@ class YOLOWrapper:
         """
         try:
             logger.info(f"Loading YOLO model from {self.model_path}")
+            # Import YOLO lazily to ensure runtime patches are applied first.
+            from ultralytics import YOLO
+
             self.model = YOLO(self.model_path)
             self.model.to(self.device)
             logger.info("Model loaded successfully")
@@ -89,6 +100,16 @@ class YOLOWrapper:
                 f"Data: {data_yaml}, Epochs: {epochs}, Batch: {batch}, ImgSz: {imgsz}"
             )
 
+            # Defaults for 16-bit safety: disable augmentations that force uint8 and HSV ops that assume 0..255.
+            # Users can override by passing explicit kwargs.
+            kwargs.setdefault("mosaic", 0.0)
+            kwargs.setdefault("mixup", 0.0)
+            kwargs.setdefault("cutmix", 0.0)
+            kwargs.setdefault("copy_paste", 0.0)
+            kwargs.setdefault("hsv_h", 0.0)
+            kwargs.setdefault("hsv_s", 0.0)
+            kwargs.setdefault("hsv_v", 0.0)
+
             # Train the model
             results = self.model.train(
                 data=data_yaml,

src/utils/image.py

@@ -313,7 +313,8 @@ class Image:
         """String representation of the Image object."""
         return (
             f"Image(path='{self.path.name}', "
-            f"shape=({self._width}x{self._height}x{self._channels}), "
+            # Display as HxWxC to match the conventional NumPy shape semantics.
+            f"shape=({self._height}x{self._width}x{self._channels}), "
             f"format={self._format}, "
             f"size={self.size_mb:.2f}MB)"
         )