Fix grayscale models tracking on videos (#20634)

This commit is contained in:
Laughing 2025-05-14 20:14:03 +08:00 committed by GitHub
parent 62180db48c
commit 1c4180832b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 31 additions and 11 deletions

View File

@ -244,9 +244,9 @@ def load_inference_source(source=None, batch=1, vid_stride=1, buffer=False, chan
elif in_memory:
dataset = source
elif stream:
dataset = LoadStreams(source, vid_stride=vid_stride, buffer=buffer)
dataset = LoadStreams(source, vid_stride=vid_stride, buffer=buffer, channels=channels)
elif screenshot:
dataset = LoadScreenshots(source)
dataset = LoadScreenshots(source, channels=channels)
elif from_img:
dataset = LoadPilAndNumpy(source, channels=channels)
else:

View File

@ -68,6 +68,7 @@ class LoadStreams:
shape (List[Tuple[int, int, int]]): List of shapes for each stream.
caps (List[cv2.VideoCapture]): List of cv2.VideoCapture objects for each stream.
bs (int): Batch size for processing.
cv2_flag (int): OpenCV flag for image reading (grayscale or RGB).
Methods:
update: Read stream frames in daemon thread.
@ -89,13 +90,14 @@ class LoadStreams:
- The class implements a buffer system to manage frame storage and retrieval.
"""
def __init__(self, sources="file.streams", vid_stride=1, buffer=False):
def __init__(self, sources="file.streams", vid_stride=1, buffer=False, channels=3):
"""Initialize stream loader for multiple video sources, supporting various stream types."""
torch.backends.cudnn.benchmark = True # faster for fixed-size inference
self.buffer = buffer # buffer input streams
self.running = True # running flag for Thread
self.mode = "stream"
self.vid_stride = vid_stride # video frame-rate stride
self.cv2_flag = cv2.IMREAD_GRAYSCALE if channels == 1 else cv2.IMREAD_COLOR # grayscale or RGB
sources = Path(sources).read_text().rsplit() if os.path.isfile(sources) else [sources]
n = len(sources)
@ -131,6 +133,7 @@ class LoadStreams:
self.fps[i] = max((fps if math.isfinite(fps) else 0) % 100, 0) or 30 # 30 FPS fallback
success, im = self.caps[i].read() # guarantee first frame
im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)[..., None] if self.cv2_flag == cv2.IMREAD_GRAYSCALE else im
if not success or im is None:
raise ConnectionError(f"{st}Failed to read images from {s}")
self.imgs[i].append(im)
@ -149,6 +152,9 @@ class LoadStreams:
cap.grab() # .read() = .grab() followed by .retrieve()
if n % self.vid_stride == 0:
success, im = cap.retrieve()
im = (
cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)[..., None] if self.cv2_flag == cv2.IMREAD_GRAYSCALE else im
)
if not success:
im = np.zeros(self.shape[i], dtype=np.uint8)
LOGGER.warning("Video stream unresponsive, please check your IP camera connection.")
@ -230,6 +236,7 @@ class LoadScreenshots:
bs (int): Batch size, set to 1.
fps (int): Frames per second, set to 30.
monitor (Dict[str, int]): Monitor configuration details.
cv2_flag (int): OpenCV flag for image reading (grayscale or RGB).
Methods:
__iter__: Returns an iterator object.
@ -241,7 +248,7 @@ class LoadScreenshots:
... print(f"Captured frame: {im.shape}")
"""
def __init__(self, source):
def __init__(self, source, channels=3):
"""Initialize screenshot capture with specified screen and region parameters."""
check_requirements("mss")
import mss # noqa
@ -259,6 +266,7 @@ class LoadScreenshots:
self.sct = mss.mss()
self.bs = 1
self.fps = 30
self.cv2_flag = cv2.IMREAD_GRAYSCALE if channels == 1 else cv2.IMREAD_COLOR # grayscale or RGB
# Parse monitor shape
monitor = self.sct.monitors[self.screen]
@ -275,6 +283,7 @@ class LoadScreenshots:
def __next__(self):
"""Captures and returns the next screenshot as a numpy array using the mss library."""
im0 = np.asarray(self.sct.grab(self.monitor))[:, :, :3] # BGRA to BGR
im0 = cv2.cvtColor(im0, cv2.COLOR_BGR2GRAY)[..., None] if self.cv2_flag == cv2.IMREAD_GRAYSCALE else im0
s = f"screen {self.screen} (LTWH): {self.left},{self.top},{self.width},{self.height}: "
self.frame += 1
@ -395,6 +404,11 @@ class LoadImagesAndVideos:
if success:
success, im0 = self.cap.retrieve()
im0 = (
cv2.cvtColor(im0, cv2.COLOR_BGR2GRAY)[..., None]
if self.cv2_flag == cv2.IMREAD_GRAYSCALE
else im0
)
if success:
self.frame += 1
paths.append(path)
@ -497,6 +511,8 @@ class LoadPilAndNumpy:
# adding new axis if it's grayscale, and converting to BGR if it's RGB
im = im[..., None] if flag == "L" else im[..., ::-1]
im = np.ascontiguousarray(im) # contiguous
elif im.ndim == 2: # grayscale in numpy form
im = im[..., None]
return im
def __len__(self):

View File

@ -330,7 +330,11 @@ class BYTETracker:
# Predict the current location with KF
self.multi_predict(strack_pool)
if hasattr(self, "gmc") and img is not None:
warp = self.gmc.apply(img, dets)
# use try-except here to bypass errors from gmc module
try:
warp = self.gmc.apply(img, dets)
except Exception:
warp = np.eye(2, 3)
STrack.multi_gmc(strack_pool, warp)
STrack.multi_gmc(unconfirmed, warp)

View File

@ -132,8 +132,8 @@ class GMC:
[[1. 0. 0.]
[0. 1. 0.]]
"""
height, width, _ = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY)
height, width, c = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY) if c == 3 else raw_frame
H = np.eye(2, 3, dtype=np.float32)
# Downscale image
@ -178,8 +178,8 @@ class GMC:
>>> print(transformation_matrix.shape)
(2, 3)
"""
height, width, _ = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY)
height, width, c = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY) if c == 3 else raw_frame
H = np.eye(2, 3)
# Downscale image
@ -320,8 +320,8 @@ class GMC:
[[1. 0. 0.]
[0. 1. 0.]]
"""
height, width, _ = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY)
height, width, c = raw_frame.shape
frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2GRAY) if c == 3 else raw_frame
H = np.eye(2, 3)
# Downscale image