feat: initial version

.gitignore

@@ -0,0 +1,156 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# Models
+*.pt

app.py

@@ -1,8 +1,22 @@
 import gradio as gr
 
+from inference import inference
 
-def greet(name):
-    return "Hello " + name + "!!"
+dd_model = gr.Dropdown(choices=["YoloV7"], value="YoloV7", label="Model")
+
+cb_motion_estimation = gr.Checkbox(value=True, label="Motion estimation")
+
+cb_path_draw = gr.Checkbox(value=True, label="Drawing paths")
+
+dd_track_points = gr.Dropdown(
+    choices=["Boxes", "Centroid"], value="Boxes", label="Detections style"
+)
+
+slide_threshold = gr.Slider(minimum=0, maximum=1, value=0.25, label="Model confidence threshold")
+
+inputs = ["video", dd_model, cb_motion_estimation, cb_path_draw, dd_track_points, slide_threshold]
+outputs = "playablevideo"
+
+iface = gr.Interface(fn=inference, inputs=inputs, outputs=outputs)
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
 iface.launch()

inference.py

@@ -0,0 +1,102 @@
+import argparse
+import glob
+import os
+
+import numpy as np
+
+from inference_utils import (
+    YOLO,
+    ModelsPath,
+    Style,
+    center,
+    clean_videos,
+    draw,
+    euclidean_distance,
+    iou,
+    yolo_detections_to_norfair_detections,
+)
+from norfair.norfair import Paths, Tracker, Video
+from norfair.norfair.camera_motion import (
+    HomographyTransformationGetter,
+    MotionEstimator,
+)
+
+DISTANCE_THRESHOLD_BBOX: float = 3.33
+DISTANCE_THRESHOLD_CENTROID: int = 30
+MAX_DISTANCE: int = 10000
+
+parser = argparse.ArgumentParser(description="Track objects in a video.")
+parser.add_argument("--img-size", type=int, default="720", help="YOLOv7 inference size (pixels)")
+parser.add_argument(
+    "--iou-threshold", type=float, default="0.45", help="YOLOv7 IOU threshold for NMS"
+)
+parser.add_argument(
+    "--classes", nargs="+", type=int, help="Filter by class: --classes 0, or --classes 0 2 3"
+)
+args = parser.parse_args()
+
+
+def inference(
+    input_video: str,
+    model: str,
+    motion_estimation: bool,
+    drawing_paths: bool,
+    track_points: str,
+    model_threshold: str,
+):
+    clean_videos("tmp")
+
+    coord_transformations = None
+    paths_drawer = None
+    track_points = Style[track_points].value
+    model = YOLO(ModelsPath[model].value, device="cuda")
+    video = Video(input_path=input_video, output_path="tmp")
+
+    if motion_estimation:
+        transformations_getter = HomographyTransformationGetter()
+
+        motion_estimator = MotionEstimator(
+            max_points=500,
+            min_distance=7,
+            transformations_getter=transformations_getter,
+            draw_flow=True,
+        )
+
+    distance_function = iou if track_points == "bbox" else euclidean_distance
+    distance_threshold = (
+        DISTANCE_THRESHOLD_BBOX if track_points == "bbox" else DISTANCE_THRESHOLD_CENTROID
+    )
+    tracker = Tracker(
+        distance_function=distance_function,
+        distance_threshold=distance_threshold,
+    )
+
+    if drawing_paths:
+        paths_drawer = Paths(center, attenuation=0.01)
+
+    for frame in video:
+        yolo_detections = model(
+            frame,
+            conf_threshold=model_threshold,
+            iou_threshold=args.iou_threshold,
+            image_size=720,
+            classes=args.classes,
+        )
+
+        mask = np.ones(frame.shape[:2], frame.dtype)
+
+        if motion_estimation:
+            coord_transformations = motion_estimator.update(frame, mask)
+
+        detections = yolo_detections_to_norfair_detections(
+            yolo_detections, track_points=track_points
+        )
+
+        tracked_objects = tracker.update(
+            detections=detections, coord_transformations=coord_transformations
+        )
+
+        frame = draw(paths_drawer, track_points, frame, detections, tracked_objects)
+        video.write(frame)
+
+    return f"{input_video[1:-4]}_out.mp4"

inference_utils.py

@@ -0,0 +1,163 @@
+import argparse
+import glob
+import os
+from enum import Enum
+from typing import List, Optional, Union
+
+import numpy as np
+import torch
+import torchvision.ops.boxes as bops
+
+from norfair import norfair
+from norfair.norfair import Detection
+
+DISTANCE_THRESHOLD_BBOX: float = 3.33
+DISTANCE_THRESHOLD_CENTROID: int = 30
+MAX_DISTANCE: int = 10000
+
+
+class ModelsPath(Enum):
+    YoloV7 = "models/yolov7.pt"
+
+
+class Style(Enum):
+    Boxes = "bbox"
+    Centroid = "centroid"
+
+
+class YOLO:
+    def __init__(self, model_path: str, device: Optional[str] = None):
+        if device is not None and "cuda" in device and not torch.cuda.is_available():
+            raise Exception("Selected device='cuda', but cuda is not available to Pytorch.")
+        # automatically set device if its None
+        elif device is None:
+            device = "cuda:0" if torch.cuda.is_available() else "cpu"
+
+        if not os.path.exists(model_path):
+            os.system(
+                f"wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/{os.path.basename(model_path)} -O {model_path}"
+            )
+
+        # load model
+        try:
+            self.model = torch.hub.load("WongKinYiu/yolov7", "custom", model_path)
+        except:
+            raise Exception("Failed to load model from {}".format(model_path))
+
+    def __call__(
+        self,
+        img: Union[str, np.ndarray],
+        conf_threshold: float = 0.25,
+        iou_threshold: float = 0.45,
+        image_size: int = 720,
+        classes: Optional[List[int]] = None,
+    ) -> torch.tensor:
+
+        self.model.conf = conf_threshold
+        self.model.iou = iou_threshold
+        if classes is not None:
+            self.model.classes = classes
+        detections = self.model(img, size=image_size)
+        return detections
+
+
+def euclidean_distance(detection, tracked_object):
+    return np.linalg.norm(detection.points - tracked_object.estimate)
+
+
+def center(points):
+    return [np.mean(np.array(points), axis=0)]
+
+
+def iou_pytorch(detection, tracked_object):
+    # Slower but simplier version of iou
+
+    detection_points = np.concatenate([detection.points[0], detection.points[1]])
+    tracked_object_points = np.concatenate([tracked_object.estimate[0], tracked_object.estimate[1]])
+
+    box_a = torch.tensor([detection_points], dtype=torch.float)
+    box_b = torch.tensor([tracked_object_points], dtype=torch.float)
+    iou = bops.box_iou(box_a, box_b)
+
+    # Since 0 <= IoU <= 1, we define 1/IoU as a distance.
+    # Distance values will be in [1, inf)
+    return np.float(1 / iou if iou else MAX_DISTANCE)
+
+
+def iou(detection, tracked_object):
+    # Detection points will be box A
+    # Tracked objects point will be box B.
+
+    box_a = np.concatenate([detection.points[0], detection.points[1]])
+    box_b = np.concatenate([tracked_object.estimate[0], tracked_object.estimate[1]])
+
+    x_a = max(box_a[0], box_b[0])
+    y_a = max(box_a[1], box_b[1])
+    x_b = min(box_a[2], box_b[2])
+    y_b = min(box_a[3], box_b[3])
+
+    # Compute the area of intersection rectangle
+    inter_area = max(0, x_b - x_a + 1) * max(0, y_b - y_a + 1)
+
+    # Compute the area of both the prediction and tracker
+    # rectangles
+    box_a_area = (box_a[2] - box_a[0] + 1) * (box_a[3] - box_a[1] + 1)
+    box_b_area = (box_b[2] - box_b[0] + 1) * (box_b[3] - box_b[1] + 1)
+
+    # Compute the intersection over union by taking the intersection
+    # area and dividing it by the sum of prediction + tracker
+    # areas - the interesection area
+    iou = inter_area / float(box_a_area + box_b_area - inter_area)
+
+    # Since 0 <= IoU <= 1, we define 1/IoU as a distance.
+    # Distance values will be in [1, inf)
+    return 1 / iou if iou else (MAX_DISTANCE)
+
+
+def yolo_detections_to_norfair_detections(
+    yolo_detections: torch.tensor, track_points: str = "centroid"  # bbox or centroid
+) -> List[Detection]:
+    """convert detections_as_xywh to norfair detections"""
+    norfair_detections: List[Detection] = []
+
+    if track_points == "centroid":
+        detections_as_xywh = yolo_detections.xywh[0]
+        for detection_as_xywh in detections_as_xywh:
+            centroid = np.array([detection_as_xywh[0].item(), detection_as_xywh[1].item()])
+            scores = np.array([detection_as_xywh[4].item()])
+            norfair_detections.append(Detection(points=centroid, scores=scores))
+    elif track_points == "bbox":
+        detections_as_xyxy = yolo_detections.xyxy[0]
+        for detection_as_xyxy in detections_as_xyxy:
+            bbox = np.array(
+                [
+                    [detection_as_xyxy[0].item(), detection_as_xyxy[1].item()],
+                    [detection_as_xyxy[2].item(), detection_as_xyxy[3].item()],
+                ]
+            )
+            scores = np.array([detection_as_xyxy[4].item(), detection_as_xyxy[4].item()])
+            norfair_detections.append(Detection(points=bbox, scores=scores))
+
+    return norfair_detections
+
+
+def clean_videos(path: str):
+    # Remove past videos
+    files = glob.glob(f"{path}/*")
+    for file in files:
+        if file.endswith(".mp4"):
+            os.remove(file)
+
+
+def draw(paths_drawer, track_points, frame, detections, tracked_objects):
+    if track_points == "centroid":
+        norfair.draw_points(frame, detections)
+        norfair.draw_tracked_objects(frame, tracked_objects)
+    elif track_points == "bbox":
+        norfair.draw_boxes(frame, detections)
+        norfair.draw_tracked_boxes(frame, tracked_objects)
+
+    if paths_drawer is not None:
+        frame = paths_drawer.draw(frame, tracked_objects)
+
+    return frame