refactor run_pipeline function for improved clarity and efficiency; add trigger class index handling and streamline detection logic
All checks were successful
Build Backend Application and Docker Image / build-docker (push) Successful in 9m27s

This commit is contained in:
Siwat Sirichai 2025-05-28 19:31:22 +07:00
parent d4754fcd27
commit f6014abb7a

View file

@ -27,10 +27,21 @@ def load_pipeline_node(node_config: dict, mpta_dir: str) -> dict:
model.to("cuda")
else:
logger.info(f"CUDA not available. Using CPU for model {node_config['modelId']}")
# Prepare trigger class indices for optimization
trigger_classes = node_config.get("triggerClasses", [])
trigger_class_indices = None
if trigger_classes and hasattr(model, "names"):
# Convert class names to indices for the model
trigger_class_indices = [i for i, name in model.names.items()
if name in trigger_classes]
logger.debug(f"Converted trigger classes to indices: {trigger_class_indices}")
node = {
"modelId": node_config["modelId"],
"modelFile": node_config["modelFile"],
"triggerClasses": node_config.get("triggerClasses", []),
"triggerClasses": trigger_classes,
"triggerClassIndices": trigger_class_indices,
"crop": node_config.get("crop", False),
"minConfidence": node_config.get("minConfidence", None),
"model": model,
@ -158,130 +169,96 @@ def load_pipeline_from_zip(zip_source: str, target_dir: str) -> dict:
logger.error(f"Error loading pipeline.json: {str(e)}", exc_info=True)
return None
def run_pipeline(frame, node: dict, return_bbox: bool = False, is_last_stage: bool = True):
def run_pipeline(frame, node: dict, return_bbox: bool=False):
"""
Processes the frame with the given pipeline node. When return_bbox is True,
the function returns a tuple (detection, bbox) where bbox is (x1,y1,x2,y2)
for drawing. Otherwise, returns only the detection.
The is_last_stage parameter controls whether this node is considered the last
in the pipeline chain. Only the last stage will return detection results.
- For detection nodes (task != 'classify'):
runs `track(..., classes=triggerClassIndices)`
picks top box minConfidence
optionally crops & resizes recurse into child
else returns (det_dict, bbox)
- For classify nodes:
runs `predict()`
returns top (class,confidence) and no bbox
"""
try:
# Check model type and use appropriate method
model_task = getattr(node["model"], "task", None)
if model_task == "classify":
# Classification models need to use predict() instead of track()
logger.debug(f"Running classification model: {node.get('modelId')}")
task = getattr(node["model"], "task", None)
# ─── Classification stage ───────────────────────────────────
if task == "classify":
# run the classifier and grab its top-1 directly via the Probs API
results = node["model"].predict(frame, stream=False)
detection = None
best_box = None
# Process classification results
for r in results:
probs = r.probs
if probs is not None and len(probs) > 0:
# Get the most confident class
class_id = int(probs.top1)
conf = float(probs.top1conf)
detection = {
"class": node["model"].names[class_id],
"confidence": conf,
"id": None # Classification doesn't have tracking IDs
}
logger.debug(f"Classification detection: {detection}")
else:
logger.debug(f"Empty classification results for model {node.get('modelId')}")
# Classification doesn't produce bounding boxes
bbox = None
else:
# Detection/segmentation models use tracking
logger.debug(f"Running detection/tracking model: {node.get('modelId')}")
results = node["model"].track(frame, stream=False, persist=True)
detection = None
best_box = None
max_conf = -1
# nothing returned?
if not results:
return (None, None) if return_bbox else None
# Log raw detection count
detection_count = 0
for r in results:
if hasattr(r.boxes, 'cpu') and len(r.boxes.cpu()) > 0:
detection_count += len(r.boxes.cpu())
if detection_count == 0:
logger.debug(f"Empty detection results (no objects found) for model {node.get('modelId')}")
else:
logger.debug(f"Detection model {node.get('modelId')} found {detection_count} objects")
# take the first result's probs object
r = results[0]
probs = r.probs
if probs is None:
return (None, None) if return_bbox else None
for r in results:
for box in r.boxes:
box_cpu = box.cpu()
conf = float(box_cpu.conf[0])
if conf > max_conf and hasattr(box, "id") and box.id is not None:
max_conf = conf
detection = {
"class": node["model"].names[int(box_cpu.cls[0])],
"confidence": conf,
"id": box.id.item()
}
best_box = box_cpu
if detection:
logger.debug(f"Best detection: {detection}")
else:
logger.debug(f"No valid detection with tracking ID for model {node.get('modelId')}")
# get the top-1 class index and its confidence
top1_idx = int(probs.top1)
top1_conf = float(probs.top1conf)
bbox = None
# Calculate bbox if best_box exists
if detection and best_box is not None:
coords = best_box.xyxy[0]
x1, y1, x2, y2 = map(int, coords)
h, w = frame.shape[:2]
x1, y1 = max(0, x1), max(0, y1)
x2, y2 = min(w, x2), min(h, y2)
if x2 > x1 and y2 > y1:
bbox = (x1, y1, x2, y2)
logger.debug(f"Detection bounding box: {bbox}")
if node.get("crop", False):
frame = frame[y1:y2, x1:x2]
logger.debug(f"Cropped frame to {frame.shape}")
det = {
"class": node["model"].names[top1_idx],
"confidence": top1_conf,
"id": None
}
return (det, None) if return_bbox else det
# ─── Detection stage ────────────────────────────────────────
# only look for your triggerClasses
tk = node["triggerClassIndices"]
res = node["model"].track(
frame,
stream=False,
persist=True,
**({"classes": tk} if tk else {})
)[0]
dets, boxes = [], []
for box in res.boxes:
conf = float(box.cpu().conf[0])
cid = int(box.cpu().cls[0])
name = node["model"].names[cid]
if conf < node["minConfidence"]:
continue
xy = box.cpu().xyxy[0]
x1,y1,x2,y2 = map(int, xy)
dets.append({"class": name, "confidence": conf,
"id": box.id.item() if hasattr(box, "id") else None})
boxes.append((x1, y1, x2, y2))
if not dets:
return (None, None) if return_bbox else None
# take highestconfidence
best_idx = max(range(len(dets)), key=lambda i: dets[i]["confidence"])
best_det = dets[best_idx]
best_box = boxes[best_idx]
# ─── Branch (classification) ───────────────────────────────
for br in node["branches"]:
if (best_det["class"] in br["triggerClasses"]
and best_det["confidence"] >= br["minConfidence"]):
# crop if requested
sub = frame
if br["crop"]:
x1,y1,x2,y2 = best_box
sub = frame[y1:y2, x1:x2]
sub = cv2.resize(sub, (224, 224))
det2, _ = run_pipeline(sub, br, return_bbox=True)
if det2:
# return classification result + original bbox
return (det2, best_box) if return_bbox else det2
# ─── No branch matched → return this detection ─────────────
return (best_det, best_box) if return_bbox else best_det
# Check if we should process branches
if detection is not None:
for branch in node["branches"]:
if detection["class"] in branch.get("triggerClasses", []):
min_conf = branch.get("minConfidence")
if min_conf is not None and detection["confidence"] < min_conf:
logger.debug(f"Confidence {detection['confidence']} below threshold {min_conf} for branch {branch['modelId']}.")
break
# If we have branches, this is not the last stage
branch_result = run_pipeline(frame, branch, return_bbox, is_last_stage=True)
# This node is no longer the last stage, so its results shouldn't be returned
is_last_stage = False
if branch_result is not None:
if return_bbox:
return branch_result
return branch_result
break
# Return this node's detection only if it's considered the last stage
if is_last_stage:
if return_bbox:
return detection, bbox
return detection
# No detection or not the last stage
if return_bbox:
return None, None
return None
except Exception as e:
logger.error(f"Error running pipeline on node {node.get('modelId')}: {e}")
if return_bbox:
return None, None
return None
logging.error(f"Error in node {node.get('modelId')}: {e}")
return (None, None) if return_bbox else None