Merge branch 'main' into fix/read_image_as_pil_chw_format

Author: onuralpszr

.github/workflows/ci.yml

@@ -39,7 +39,7 @@ jobs:
     steps:
       - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
       - name: Setup uv python package manager
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
         with:
           python-version: ${{ matrix.python-version }}
           enable-cache: true
@@ -96,7 +96,7 @@ jobs:
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
     steps:
       - name: Setup uv python package manager
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
         with:
           python-version: ${{ matrix.python-version }}
           enable-cache: true

.github/workflows/mmdet.yml

@@ -69,7 +69,7 @@ jobs:
     steps:
       - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
       - name: Setup uv python package manager
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
         with:
           python-version: ${{ matrix.python-version }}
           enable-cache: true

.github/workflows/publish_docs.yml

@@ -34,7 +34,7 @@ jobs:
           fetch-depth: 0
 
       - name: 🐍 Install uv and set Python ${{ matrix.python-version }}
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
         with:
           python-version: ${{ matrix.python-version }}
           activate-environment: true

.github/workflows/publish_pypi.yml

@@ -10,7 +10,7 @@ jobs:
     steps:
       - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
       - name: Setup uv python package manager
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
         with:
           enable-cache: true
           prune-cache: false

.github/workflows/ruff.yaml

@@ -26,7 +26,7 @@ jobs:
         with:
           python-version: '3.10'
       - name: Install uv
-        uses: astral-sh/setup-uv@85856786d1ce8acfbcc2f13a5f3fbd6b938f9f41 # v7.1.2
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
       - name: Install dependencies
         run: |
           uv venv --python 3.10

demo/inference_for_ultralytics_yoloe.ipynb

@@ -56,6 +56,7 @@ nav:
       - Inference RTDETR: notebooks/inference_for_rtdetr.ipynb
       - Inference Roboflow: notebooks/inference_for_roboflow.ipynb
       - Inference TorchVision: notebooks/inference_for_torchvision.ipynb
+      - Inference YOLOE: notebooks/inference_for_ultralytics_yoloe.ipynb
       - Slicing: notebooks/slicing.ipynb
 
   - Code Reference:

mkdocs.yml

@@ -26,7 +26,7 @@ class YOLOEDetectionModel(UltralyticsDetectionModel):
             - yoloe-11s-seg-pf.pt, yoloe-11m-seg-pf.pt, yoloe-11l-seg-pf.pt
             - yoloe-v8s-seg-pf.pt, yoloe-v8m-seg-pf.pt, yoloe-v8l-seg-pf.pt
 
-    !!! example "Usage with text prompts"
+    !!! example "Usage Text Prompts"
         ```python
         from sahi import AutoDetectionModel
 

sahi/models/yoloe.py

@@ -1,8 +1,8 @@
 from __future__ import annotations
 
+import numpy as np
 import torch
-from shapely.geometry import box
-from shapely.strtree import STRtree
+from shapely import STRtree, box
 
 from sahi.logger import logger
 from sahi.postprocess.utils import ObjectPredictionList, has_match, merge_object_prediction_pair
@@ -53,31 +53,30 @@ def nms(
     Returns:
         A list of filtered indexes, Shape: [ ,]
     """
+    if len(predictions) == 0:
+        return []
 
-    # Extract coordinates and scores as tensors
-    x1 = predictions[:, 0]
-    y1 = predictions[:, 1]
-    x2 = predictions[:, 2]
-    y2 = predictions[:, 3]
-    scores = predictions[:, 4]
+    # Ensure predictions are on CPU and convert to numpy
+    if predictions.device.type != "cpu":
+        predictions = predictions.cpu()
 
-    # Calculate areas as tensor (vectorized operation)
+    predictions_np = predictions.numpy()
+
+    # Extract coordinates and scores
+    x1 = predictions_np[:, 0]
+    y1 = predictions_np[:, 1]
+    x2 = predictions_np[:, 2]
+    y2 = predictions_np[:, 3]
+    scores = predictions_np[:, 4]
+
+    # Calculate areas
     areas = (x2 - x1) * (y2 - y1)
 
-    # Create Shapely boxes only once
-    boxes = []
-    for i in range(len(predictions)):
-        boxes.append(
-            box(
-                x1[i].item(),  # Convert only individual values
-                y1[i].item(),
-                x2[i].item(),
-                y2[i].item(),
-            )
-        )
+    # Create Shapely boxes (vectorized)
+    boxes = box(x1, y1, x2, y2)
 
-    # Sort indices by score (descending) using torch
-    sorted_idxs = torch.argsort(scores, descending=True).tolist()
+    # Sort indices by score (descending)
+    sorted_idxs = np.argsort(scores)[::-1]
 
     # Build STRtree
     tree = STRtree(boxes)
@@ -91,7 +90,7 @@ def nms(
 
         keep.append(current_idx)
         current_box = boxes[current_idx]
-        current_area = areas[current_idx].item()  # Convert only when needed
+        current_area = areas[current_idx]
 
         # Query potential intersections using STRtree
         candidate_idxs = tree.query(current_box)
@@ -108,16 +107,16 @@ def nms(
             if scores[candidate_idx] == scores[current_idx]:
                 # Use box coordinates for stable ordering
                 current_coords = (
-                    x1[current_idx].item(),
-                    y1[current_idx].item(),
-                    x2[current_idx].item(),
-                    y2[current_idx].item(),
+                    x1[current_idx],
+                    y1[current_idx],
+                    x2[current_idx],
+                    y2[current_idx],
                 )
                 candidate_coords = (
-                    x1[candidate_idx].item(),
-                    y1[candidate_idx].item(),
-                    x2[candidate_idx].item(),
-                    y2[candidate_idx].item(),
+                    x1[candidate_idx],
+                    y1[candidate_idx],
+                    x2[candidate_idx],
+                    y2[candidate_idx],
                 )
 
                 # Compare coordinates lexicographically
@@ -130,10 +129,10 @@ def nms(
 
             # Calculate metric
             if match_metric == "IOU":
-                union = current_area + areas[candidate_idx].item() - intersection
+                union = current_area + areas[candidate_idx] - intersection
                 metric = intersection / union if union > 0 else 0
             elif match_metric == "IOS":
-                smaller = min(current_area, areas[candidate_idx].item())
+                smaller = min(current_area, areas[candidate_idx])
                 metric = intersection / smaller if smaller > 0 else 0
             else:
                 raise ValueError("Invalid match_metric")

sahi/postprocess/combine.py

@@ -63,8 +63,12 @@ def get_slice_bboxes(
     y_max = y_min = 0
 
     if slice_height and slice_width:
-        y_overlap = int(overlap_height_ratio * slice_height)
-        x_overlap = int(overlap_width_ratio * slice_width)
+        if overlap_height_ratio is not None and overlap_height_ratio >= 1.0:
+            raise ValueError("Overlap ratio must be less than 1.0")
+        if overlap_width_ratio is not None and overlap_width_ratio >= 1.0:
+            raise ValueError("Overlap ratio must be less than 1.0")
+        y_overlap = int((overlap_height_ratio if overlap_height_ratio is not None else 0.2) * slice_height)
+        x_overlap = int((overlap_width_ratio if overlap_width_ratio is not None else 0.2) * slice_width)
     elif auto_slice_resolution:
         x_overlap, y_overlap, slice_width, slice_height = get_auto_slice_params(height=image_height, width=image_width)
     else: