feat: per-class autolabel params + frame-scoped class clear (REQ-180, REQ-181)

- REQ-181: class_params {name: {threshold?, iou_threshold?, min_box_frac?}}
  on /preview and /autolabel, per-class override table in both modals;
  empty overrides take the unchanged global path
- REQ-180: x button on each review sidebar class row clears that class on
  the current frame only via bulk-delete, no confirmation
- includes REQ-178 empty date-folder cycle fix (archive_index.py)

backend/api/batches.py

@@ -35,6 +35,7 @@ class AutolabelRequest(BaseModel):
     resume: bool = False
     append: bool = False
     custom_model_path: Optional[str] = None
+    class_params: Optional[dict[str, dict[str, float]]] = None
 
 class Exemplar(BaseModel):
     box: list[float]          # [cx, cy, w, h], normalized 0..1
@@ -53,6 +54,7 @@ class PreviewRequest(BaseModel):
     # Preview-only — `autolabel.start` deliberately has no equivalent.
     exemplars: Optional[list[Exemplar]] = None
     exemplar_class_name: Optional[str] = None
+    class_params: Optional[dict[str, dict[str, float]]] = None
 
 
 @router.post("/api/projects/{project_id}/batches")
@@ -116,7 +118,8 @@ def start_autolabel(batch_id: int, request: AutolabelRequest) -> dict:
                                engine=request.engine, engines=engine_list, class_ids=request.class_ids,
                                engine_classes=request.engine_classes,
                                target_class_names=request.target_class_names,
-                               custom_model_path=request.custom_model_path)
+                               custom_model_path=request.custom_model_path,
+                               class_params=request.class_params)
     except batch_store.BatchError as exc:
         raise HTTPException(400, str(exc))
 @router.post("/api/batches/inspect-model")
@@ -189,6 +192,7 @@ def preview_autolabel(batch_id: int, request: PreviewRequest) -> dict:
             custom_model_path=request.custom_model_path,
             exemplars=[e.model_dump() for e in request.exemplars or []],
             exemplar_class_name=request.exemplar_class_name,
+            class_params=request.class_params,
         )
         return {"shapes": shapes}
     except Exception as exc:

backend/archive_index.py

@@ -167,7 +167,14 @@ def cycles(project_id: int) -> List[dict]:
 
     buckets: dict = {}
     for day in library.list_dates(project["video_root"]):
-        for name in _video_names(project, day["date"]):
+        names = _video_names(project, day["date"])
+        if not names:
+            # Folder still holds no recording: it must still appear in the
+            # list, or a folder just created with "Folder baru" is invisible.
+            buckets.setdefault(day["date"], {"cycle": day["date"], "video_count": 0,
+                                             "flagged": 0, "first_start": None})
+            continue
+        for name in names:
             rel = f"{day['date']}/{name}"
             timing = _sidecar(project, rel) or known.get(rel) or {}
             cycle = timing.get("working_day") or day["date"]

backend/autolabel.py

@@ -17,6 +17,26 @@ DEFAULT_THRESHOLD = 0.35
 DEFAULT_IOU = 0.0
 
 
+def _parse_class_params(raw) -> dict:
+    """Normalize `class_params` (REQ-181): lowercased class name → overrides.
+
+    Unknown keys and non-numeric values are dropped, so a malformed payload
+    degrades to the global values instead of failing the job."""
+    out: dict = {}
+    for name, values in (raw or {}).items():
+        if not isinstance(values, dict):
+            continue
+        entry = {}
+        for key in ("threshold", "iou_threshold", "min_box_frac"):
+            if key in values:
+                try:
+                    entry[key] = float(values[key])
+                except (TypeError, ValueError):
+                    pass
+        if entry:
+            out[str(name).strip().lower()] = entry
+    return out
+
 
 
 def start(batch_id: int, threshold: float = DEFAULT_THRESHOLD,
@@ -26,7 +46,8 @@ def start(batch_id: int, threshold: float = DEFAULT_THRESHOLD,
           class_ids: Optional[List[int]] = None,
           engine_classes: Optional[dict[str, List[str]]] = None,
           custom_model_path: Optional[str] = None,
-          target_class_names: Optional[List[str]] = None) -> dict:
+          target_class_names: Optional[List[str]] = None,
+          class_params: Optional[dict] = None) -> dict:
     batch = batches.get(batch_id)
     if batch is None:
         raise batches.BatchError("No such batch")
@@ -41,7 +62,8 @@ def start(batch_id: int, threshold: float = DEFAULT_THRESHOLD,
                 "iou_threshold": iou_threshold, "min_box_frac": min_box_frac,
                 "resume": resume, "append": append, "engine": active_engines[0], "engines": active_engines,
                 "class_ids": class_ids, "engine_classes": engine_classes,
-                "custom_model_path": custom_model_path, "target_class_names": target_class_names},
+                "custom_model_path": custom_model_path, "target_class_names": target_class_names,
+                "class_params": class_params},
         project_id=batch["project_id"],
         batch_id=batch_id,
         message=f"{batch['date_label']}/{batch['batch_label']} ({'+'.join(e.upper() for e in active_engines)})",
@@ -85,6 +107,10 @@ def _run_autolabel(job) -> None:
 
     conf = job.params.get("threshold", DEFAULT_THRESHOLD)
     iou_thresh = job.params.get("iou_threshold", DEFAULT_IOU)
+    per_class = _parse_class_params(job.params.get("class_params"))
+    # Predict at the LOWEST threshold in play so a class with a lower override
+    # can still see its boxes; each box then passes its own class gate below.
+    predict_conf = min([conf] + [v["threshold"] for v in per_class.values() if "threshold" in v])
 
     yolo_model = None
     sam3_target_classes = []
@@ -159,7 +185,7 @@ def _run_autolabel(job) -> None:
             all_raw_detections = []
 
             if yolo_model is not None:
-                results = yolo_model.predict(frame_file, conf=conf, verbose=False)
+                results = yolo_model.predict(frame_file, conf=predict_conf, verbose=False)
                 if results and len(results) > 0:
                     model_names = results[0].names
                     for box in results[0].boxes:
@@ -189,6 +215,15 @@ def _run_autolabel(job) -> None:
 
                         score = float(box.conf[0].item())
                         xyxyn = box.xyxyn[0].tolist()
+
+                        cp = per_class.get(proj_cls_name) or per_class.get(raw_cls_name)
+                        if cp:
+                            if score < cp.get("threshold", conf):
+                                continue
+                            mb = cp.get("min_box_frac")
+                            if mb and mb > 0 and (xyxyn[2]-xyxyn[0])*(xyxyn[3]-xyxyn[1]) < mb:
+                                continue
+
                         all_raw_detections.append(labeling.Detection(
                             class_id=target_class_id,
                             class_name=proj_cls_name or raw_cls_name,
@@ -199,9 +234,21 @@ def _run_autolabel(job) -> None:
 
             if selected_engine == "sam3" and sam3_target_classes:
                 prompts = [(c.get("prompt") or c["name"]).strip() for c in sam3_target_classes]
+                thr_list = iou_list = mb_list = None
+                if per_class:
+                    names = [c["name"].strip().lower() for c in sam3_target_classes]
+                    if any("threshold" in v for v in per_class.values()):
+                        thr_list = [per_class.get(n, {}).get("threshold", conf) for n in names]
+                    if any("iou_threshold" in v for v in per_class.values()):
+                        iou_list = [per_class.get(n, {}).get("iou_threshold", iou_thresh) for n in names]
+                    if any("min_box_frac" in v for v in per_class.values()):
+                        mb_list = [per_class.get(n, {}).get("min_box_frac", job.params.get("min_box_frac", 0.0)) for n in names]
                 res = labeling.label_image(
                     frame_file, frame["filename"], prompts, conf,
-                    iou_threshold=iou_thresh, min_box_frac=job.params.get("min_box_frac", 0.0)
+                    iou_threshold=iou_thresh, min_box_frac=job.params.get("min_box_frac", 0.0),
+                    thresholds=thr_list,
+                    iou_by_class=dict(enumerate(iou_list)) if iou_list else None,
+                    min_box_fracs=mb_list,
                 )
                 if not res.error and res.detections:
                     for det in res.detections:
@@ -213,7 +260,14 @@ def _run_autolabel(job) -> None:
                 elif res.error:
                     job.log(f"[SAM3 ERROR] {frame['filename']}: {res.error}")
 
-            kept = labeling.deduplicate(all_raw_detections, iou_threshold=iou_thresh)
+            iou_by_class_proj = {
+                c["class_id"]: per_class[c["name"].strip().lower()]["iou_threshold"]
+                for c in project["classes"]
+                if c["name"].strip().lower() in per_class
+                and "iou_threshold" in per_class[c["name"].strip().lower()]
+            }
+            kept = labeling.deduplicate(all_raw_detections, iou_threshold=iou_thresh,
+                                        iou_by_class=iou_by_class_proj or None)
             items = []
             for det in kept:
                 if project["label_type"] == "bbox" or det.mask is None:

backend/labeling.py

@@ -10,7 +10,7 @@ calls — see the domain invariants in `../AGENTS.md`.
 """
 
 from dataclasses import dataclass, field
-from typing import List, Optional
+from typing import Dict, List, Optional
 
 from PIL import Image
 
@@ -41,9 +41,12 @@ def _iou(box_a: List[float], box_b: List[float]) -> float:
     return inter / union if union > 0 else 0.0
 
 
-def deduplicate(detections: List[Detection], iou_threshold: float = 0.8) -> List[Detection]:
-    """Greedy NMS per class: highest score wins within the SAME class."""
-    if iou_threshold <= 0.0:
+def deduplicate(detections: List[Detection], iou_threshold: float = 0.8,
+                iou_by_class: Optional[Dict[int, float]] = None) -> List[Detection]:
+    """Greedy NMS per class: highest score wins within the SAME class.
+
+    `iou_by_class` overrides the threshold per class id (REQ-181)."""
+    if not iou_by_class and iou_threshold <= 0.0:
         return detections
 
     by_class: dict[int, List[Detection]] = {}
@@ -52,10 +55,14 @@ def deduplicate(detections: List[Detection], iou_threshold: float = 0.8) -> List
         by_class.setdefault(det.class_id, []).append(det)
 
     kept: List[Detection] = []
-    for cls_dets in by_class.values():
+    for cls, cls_dets in by_class.items():
+        iou = (iou_by_class or {}).get(cls, iou_threshold)
+        if iou <= 0.0:
+            kept.extend(cls_dets)
+            continue
         cls_kept: List[Detection] = []
         for det in sorted(cls_dets, key=lambda d: d.score, reverse=True):
-            if all(_iou(det.box, k.box) < iou_threshold for k in cls_kept):
+            if all(_iou(det.box, k.box) < iou for k in cls_kept):
                 cls_kept.append(det)
         kept.extend(cls_kept)
     return kept
@@ -70,11 +77,16 @@ def label_image(
     min_box_frac: float = 0.0,
     exemplar_index: int = -1,
     exemplars: Optional[List[dict]] = None,
+    thresholds: Optional[List[float]] = None,
+    iou_by_class: Optional[Dict[int, float]] = None,
+    min_box_fracs: Optional[List[float]] = None,
 ) -> ImageResult:
     """Detect every prompt in one image and return the surviving instances.
 
     When `exemplars` are given, the prompt at `exemplar_index` also carries them
-    as drawn box exemplars (REQ-172); every other prompt runs on text alone."""
+    as drawn box exemplars (REQ-172); every other prompt runs on text alone.
+    `thresholds`, `iou_by_class` and `min_box_fracs` are per-prompt overrides
+    aligned with `prompts` (REQ-181); classes without one use the global values."""
     try:
         image = Image.open(image_path).convert("RGB")
     except Exception as exc:  # unreadable/corrupt frame: report, don't abort the job
@@ -84,14 +96,24 @@ def label_image(
     try:
         if exemplars and 0 <= exemplar_index < len(prompts):
             detections = get_engine().detect_with_exemplars(
-                image, prompts, threshold, exemplar_index, exemplars
+                image, prompts, threshold, exemplar_index, exemplars,
+                thresholds=thresholds
             )
         else:
-            detections = get_engine().detect(image, prompts, threshold)
+            detections = get_engine().detect(image, prompts, threshold,
+                                             thresholds=thresholds)
     except Exception as exc:
         return ImageResult(image_path, rel_path, width, height, error=str(exc))
 
-    if min_box_frac > 0:
+    if min_box_fracs is not None:
+        def _keep(det) -> bool:
+            frac = min_box_fracs[det.class_id] if det.class_id < len(min_box_fracs) else min_box_frac
+            if frac <= 0:
+                return True
+            floor = width * height * frac
+            return (det.box[2] - det.box[0]) * (det.box[3] - det.box[1]) >= floor
+        detections = [d for d in detections if _keep(d)]
+    elif min_box_frac > 0:
         floor = width * height * min_box_frac
         detections = [
             d for d in detections
@@ -99,4 +121,4 @@ def label_image(
         ]
 
     return ImageResult(image_path, rel_path, width, height,
-                       deduplicate(detections, iou_threshold))
+                       deduplicate(detections, iou_threshold, iou_by_class=iou_by_class))

backend/preview.py

@@ -12,7 +12,7 @@ import os
 from typing import List, Optional
 
 from backend import batches, db, labeling, projects, review
-from backend.autolabel import DEFAULT_IOU, DEFAULT_THRESHOLD, _geometries
+from backend.autolabel import DEFAULT_IOU, DEFAULT_THRESHOLD, _geometries, _parse_class_params
 
 def preview_frame(
     batch_id: int,
@@ -25,6 +25,7 @@ def preview_frame(
     custom_model_path: Optional[str] = None,
     exemplars: Optional[List[dict]] = None,
     exemplar_class_name: Optional[str] = None,
+    class_params: Optional[dict] = None,
 ) -> List[dict]:
     batch = batches.get(batch_id)
     if not batch:
@@ -69,11 +70,13 @@ def preview_frame(
 
     name_to_class_id = {item["name"].strip().lower(): item["class_id"] for item in project["classes"]}
     allowed_classes_set = {c.strip().lower() for c in target_class_names} if target_class_names else None
+    per_class = _parse_class_params(class_params)
+    predict_conf = min([threshold] + [v["threshold"] for v in per_class.values() if "threshold" in v])
 
     all_raw_detections = []
 
     if yolo_model is not None:
-        results = yolo_model.predict(frame_file, conf=threshold, verbose=False)
+        results = yolo_model.predict(frame_file, conf=predict_conf, verbose=False)
         if results and len(results) > 0:
             model_names = results[0].names
             for box in results[0].boxes:
@@ -103,6 +106,15 @@ def preview_frame(
 
                 score = float(box.conf[0].item())
                 xyxyn = box.xyxyn[0].tolist()
+
+                cp = per_class.get(proj_cls_name) or per_class.get(raw_cls_name)
+                if cp:
+                    if score < cp.get("threshold", threshold):
+                        continue
+                    mb = cp.get("min_box_frac")
+                    if mb and mb > 0 and (xyxyn[2]-xyxyn[0])*(xyxyn[3]-xyxyn[1]) < mb:
+                        continue
+
                 all_raw_detections.append(labeling.Detection(
                     class_id=target_class_id,
                     class_name=proj_cls_name or raw_cls_name,
@@ -125,10 +137,22 @@ def preview_frame(
                  if c["name"].strip().lower() == wanted),
                 -1,
             )
+        thr_list = iou_list = mb_list = None
+        if per_class:
+            names = [c["name"].strip().lower() for c in sam3_target_classes]
+            if any("threshold" in v for v in per_class.values()):
+                thr_list = [per_class.get(n, {}).get("threshold", threshold) for n in names]
+            if any("iou_threshold" in v for v in per_class.values()):
+                iou_list = [per_class.get(n, {}).get("iou_threshold", iou_threshold) for n in names]
+            if any("min_box_frac" in v for v in per_class.values()):
+                mb_list = [per_class.get(n, {}).get("min_box_frac", min_box_frac) for n in names]
         res = labeling.label_image(
             frame_file, frame["filename"], prompts, threshold,
             iou_threshold=iou_threshold, min_box_frac=min_box_frac,
-            exemplar_index=exemplar_index, exemplars=exemplars
+            exemplar_index=exemplar_index, exemplars=exemplars,
+            thresholds=thr_list,
+            iou_by_class=dict(enumerate(iou_list)) if iou_list else None,
+            min_box_fracs=mb_list,
         )
         if not res.error and res.detections:
             for det in res.detections:
@@ -138,7 +162,14 @@ def preview_frame(
                     det.class_name = real_cls["name"]
                     all_raw_detections.append(det)
 
-    kept = labeling.deduplicate(all_raw_detections, iou_threshold=iou_threshold)
+    iou_by_class_proj = {
+        c["class_id"]: per_class[c["name"].strip().lower()]["iou_threshold"]
+        for c in project["classes"]
+        if c["name"].strip().lower() in per_class
+        and "iou_threshold" in per_class[c["name"].strip().lower()]
+    }
+    kept = labeling.deduplicate(all_raw_detections, iou_threshold=iou_threshold,
+                                iou_by_class=iou_by_class_proj or None)
     items = []
     for det in kept:
         if project["label_type"] == "bbox" or det.mask is None:

backend/sam3_engine.py

@@ -67,8 +67,12 @@ class Sam3Engine:
         )
         self.processor = Sam3Processor(self.model, device=self.device)
 
-    def detect(self, image: Image.Image, prompts: List[str], threshold: float) -> List[Detection]:
-        """Run every prompt against one image; prompt index becomes the class id."""
+    def detect(self, image: Image.Image, prompts: List[str], threshold: float,
+               thresholds: Optional[List[float]] = None) -> List[Detection]:
+        """Run every prompt against one image; prompt index becomes the class id.
+
+        `thresholds` overrides the confidence per prompt (REQ-181); still one
+        `set_image` for the whole call — only the grounding head sees the change."""
         processor = Sam3Processor(self.model, device=self.device)
         processor.confidence_threshold = threshold
 
@@ -76,6 +80,8 @@ class Sam3Engine:
         with torch.autocast(self.device, dtype=self.autocast_dtype):
             state = processor.set_image(image)
             for class_id, prompt in enumerate(prompts):
+                if thresholds is not None and class_id < len(thresholds):
+                    processor.confidence_threshold = thresholds[class_id]
                 output = processor.set_text_prompt(prompt=prompt, state=state)
                 masks, boxes, scores = output["masks"], output["boxes"], output["scores"]
                 if masks.shape[0] == 0:
@@ -110,6 +116,7 @@ class Sam3Engine:
         threshold: float,
         exemplar_index: int,
         exemplars: List[dict],
+        thresholds: Optional[List[float]] = None,
     ) -> List[Detection]:
         """`detect()`, but one prompt also carries drawn box exemplars (REQ-172).
 
@@ -124,6 +131,8 @@ class Sam3Engine:
         with torch.autocast(self.device, dtype=self.autocast_dtype):
             state = processor.set_image(image)
             for class_id, prompt in enumerate(prompts):
+                if thresholds is not None and class_id < len(thresholds):
+                    processor.confidence_threshold = thresholds[class_id]
                 processor.reset_all_prompts(state)
                 output = processor.set_text_prompt(prompt=prompt, state=state)
                 if class_id == exemplar_index:

docs/design.md

@@ -174,11 +174,14 @@ GET    /api/projects/{id}/video?rel=…           # Range streaming (REQ-013)
 POST   /api/projects/{id}/batches               # {rel, start_sec, end_sec, fps} → extract job
 GET    /api/batches/{id}                        # status + review progress (REQ-045)
 GET    /api/batches/{id}/frames                 # frames + statuses
-POST   /api/batches/{id}/autolabel              # {threshold} → job (REQ-030,032,034)
+POST   /api/batches/{id}/autolabel              # {threshold, class_params} → job (REQ-030,032,034,181)
 POST   /api/batches/{id}/preview                # one frame, run now, nothing written;
+                                                #   +class_params {name: {threshold?, iou_threshold?, min_box_frac?}}
+                                                #     empty override = global value, keys are class names (REQ-181)
                                                 #   +exemplars[] {box:[cx,cy,w,h], positive}
                                                 #   +exemplar_class_name (REQ-171,172)
-DELETE /api/batches/{id}/classes/{class_id}/annotations # clear all shapes of class in batch (REQ-046)
+DELETE /api/batches/{id}/classes/{class_id}/annotations # clear all shapes of class in batch (REQ-046);
+                                                #   frame-scoped clear = client filters this frame → bulk-delete (REQ-180)
 POST   /api/batches/{ids}/approve               # one or many, comma-separated → one merge job (REQ-131)
 GET    /api/batches/{ids}/triage/summary        # one or many, comma-separated (REQ-130)
 GET    /api/batches/{ids}/triage/shapes
@@ -190,6 +193,7 @@ GET    /api/frames/{id}/annotations
 POST   /api/frames/{id}/annotations             # add a manual shape (REQ-042)
 PATCH  /api/annotations/{id}                    # move/resize/reclass
 DELETE /api/annotations/{id}
+POST   /api/annotations/bulk-delete             # {ids[]} → N deletes (REQ-180 frame-clear uses it)
 POST   /api/frames/{id}/assist                  # click/box → SAM3 shape (REQ-043)
 POST   /api/frames/{id}/exemplar-label          # drawn pool → re-detect one class (REQ-173,174)
 POST   /api/frames/{id}/status                  # approved | rejected | pending (REQ-041)
@@ -238,7 +242,13 @@ updated. Range and fps live on the batch, so one video can be used repeatedly.
 
 **autolabel (REQ-030…034).** Per frame: one `set_image`, then loop each class's prompt (see
 the domain invariants in `../AGENTS.md`), cross-prompt NMS, write `annotations` rows with
-`source='auto'`. A re-run deletes only `source='auto'` rows — manual corrections
+`source='auto'`. `class_params` (REQ-181) is consulted per class: a numeric
+`threshold` / `iou_threshold` / `min_box_frac` entry replaces the job's global for that class
+only; classes without an entry — and a run with `class_params` empty or absent — take the
+exact global path. The YOLO confidence floor uses `min([global] + overrides)`, the SAM3
+threshold/dedup/min-box lists are built only from classes that actually override that key,
+and `/preview` applies the same overrides so the preview and the job cannot disagree.
+A re-run deletes only `source='auto'` rows — manual corrections
 (`source='manual'`) survive — and returns already-approved frames to `pending`, because
 that approval was given against labels that no longer exist. No overlay images are written:
 the review canvas draws the shapes from the annotation rows, so a second rendering of the

docs/requirements.md

@@ -74,7 +74,8 @@ changes.
 
 - **REQ-178** — The Video Archive page can create a `YYYY-MM-DD` folder and upload videos
   into it. Uploads are streamed to disk, never overwrite an existing file, and only accept
-  `video.VIDEO_EXTS`.
+  `video.VIDEO_EXTS`. The new folder appears in the archive's cycle list immediately, even
+  before it holds any video.
 - **REQ-179** — The Video Archive page shows a button that copies the archive's host path
   to the clipboard, as a Linux path and as a Windows (`\\wsl.localhost\<distro>\...`) path
   when known.
@@ -97,6 +98,9 @@ changes.
 - **REQ-031** — Detections that overlap across prompts are deduplicated (greedy IoU NMS), so
   one object is not labelled as two classes at once.
 - **REQ-032** — The confidence threshold is configurable per job.
+- **REQ-181** — Auto-annotation accepts per-class overrides of confidence, IoU and minimum box
+  fraction on top of the job's global values; classes without an override use the global
+  values. Preview and the job apply the same overrides.
 - **REQ-033** — A frame with no detections is valid and still enters the dataset as a
   negative sample — it is not a failure.
 - **REQ-034** — Auto-annotation can be re-run on the same batch; previous automatic results
@@ -167,6 +171,9 @@ changes.
   auto-annotate modal, and reset with the frame. Defaults are confidence `0.5`, NMS `0.8`,
   min box `0.002`, max `100` — deliberately permissive, because on a dense frame an
   aggressive NMS or area floor deletes real, touching objects rather than duplicates.
+- **REQ-180** — In the manual review editor, each class row can clear that class's shapes on
+  the **current frame only** (REQ-046 stays batch-wide). One click, no confirmation dialog;
+  the next frame is untouched.
 
 ## E4. Live counting preview
 

docs/tasks.md

@@ -1216,6 +1216,19 @@ failure modes to the counting path. The saving was always on the browser side.
    `vite build` passes (66 modules). Browser click-test of copy buttons/upload dialog is NOT
    automated — manual click-test pending.
 
+## Task — Empty date folder visible in cycle list (REQ-178) `[DONE]`
+
+1. `archive_index.cycles()` must return a bucket for a date folder with no videos →
+   verify: **[DONE]** `POST .../library/dates?date=2099-01-01` → 200; then
+   `GET .../archive/cycles` contains `2099-01-01` with `video_count: 0` (and the
+   pre-existing empty folder `2026-08-31`, previously invisible, also appears);
+   `GET .../archive/cycles/2099-01-01` returns `videos: []`; the same JSON arrives
+   through the dev proxy on 5173; folder removed after verification → cycle gone from
+   the list again; `uvx ruff check` output identical to HEAD (13 pre-existing, 0 new);
+   `vite build` passes. Browser click-test is NOT automated — manual click-test pending.
+   Docker backend rebuilt after the fix: `:9010/.../archive/cycles` now returns the
+   empty folders `2026-08-29` and `2026-08-31` with `video_count: 0`.
+
 ## Task — Copy-path buttons (REQ-179) `[DONE]`
 
 1. Copy-path buttons → verify: **[DONE]** project payload carries
@@ -1229,6 +1242,32 @@ failure modes to the counting path. The saving was always on the browser side.
    nginx `client_max_body_size 20g` verified (600MB upload → 200); `/api/health` 200, `/docs`
    200, frontend `http://localhost:9000` 200 after rebuild.
 
+## Task — Frame-scoped per-class clear in review editor (REQ-180) `[DONE]`
+
+1. `×` button on each sidebar class row clears that class's shapes on the **current frame
+   only**, no confirmation, batch-wide trash (REQ-046) unchanged →
+   verify: **[DONE]** `ReviewPage.clearClassInFrame` filters the current frame's annotations
+   and posts only those ids to `POST /annotations/bulk-delete`; round-trip on `:9010` —
+   created 2 shapes on frame 23827, bulk-delete returned `{"deleted":2}`, frame back to
+   `annotations: []`, DB clean after test; optimistic update + rollback wired the same way as
+   `removeMarked`; `vite build` passes (68 modules), rebuilt image on `:9000` serves the new
+   bundle (`index-wBeBdXus.js`). Browser click-test of the `×` button is NOT automated —
+   manual click-test pending.
+
+## Task — Per-class auto-annotate params (REQ-181) `[DONE]`
+
+1. `class_params` through preview + job on both modals → verify: **[DONE]** ruff on the 5
+   changed backend files vs HEAD: +17, all `UP006`/`UP035`/`UP045` (the files' existing
+   style), 0 new real findings; on rebuilt backend `:9010` — `/preview` baseline (absent and
+   `null` `class_params`) → 18 shapes, `class_params: {sack: {threshold: 0.99}}` → 0 shapes,
+   `threshold: "high"` → 422 `float_parsing`; `/autolabel` accepted
+   `{class_params: {sack: {threshold: 0.9, min_box_frac: 0.01}}}`, job 106's stored params
+   carry it, `iou_threshold: "x"` → 422, job cancelled and test data reset
+   (`reset-auto-annotations`, batch 21 back to 0 annotations); omitted `class_params` takes
+   the unchanged global path; `vite build` passes, `:9000` serves the new bundle (marker
+   strings present). Browser click-test of the override tables is NOT automated — manual
+   click-test pending.
+
 ## Known open points
 
 - *Not closed by any task, by choice:* **any rebuild kills the running job.** Task 14's resume

docs/ui-spec.md

@@ -449,6 +449,11 @@ Two columns inside one dialog (920 px wide, max 96 vw / 90 vh, scrollable).
 - Confidence threshold — 0.05…0.95 step 0.05, default 0.35.
 - NMS IoU threshold — 0…0.9 step 0.05, default 0.0.
 - Min box size (fraction of frame) — 0…0.5 step 0.005, default 0, shown as a percentage.
+- **Per-class overrides** (REQ-181) — a compact table below the sliders, one row per currently
+  selected class: `Conf` / `IoU` / `MinBox` number inputs. Each input's placeholder shows the
+  current global value and an empty input inherits it; only filled cells are sent, as
+  `class_params` on **both** the preview and the job request, so preview and run cannot
+  disagree. Classes without an override behave exactly as before.
 - `ClassPromptPanel`:
   - Target-class chips. Without SAM3 a chip is a simple toggle.
   - **With SAM3 a chip is two buttons**: the *name* selects-and-activates, the `×` deselects.
@@ -470,6 +475,7 @@ preview is the answer, so no button press sits between them. Re-running is cheap
 ```json
 { "frame_id": 12, "engine": "sam3|base_model|custom",
   "threshold": 0.35, "iou_threshold": 0.0, "min_box_frac": 0.0,
+  "class_params": { "sack": { "threshold": 0.55 } },
   "target_class_names": ["sack"],
   "custom_model_path": null,
   "exemplars": [ { "box": [cx, cy, w, h], "positive": true } ],
@@ -485,10 +491,14 @@ the prompt, they are not labels.
 ```json
 { "resume": false, "append": true, "engine": "sam3",
   "threshold": 0.35, "iou_threshold": 0.0, "min_box_frac": 0.0,
+  "class_params": { "sack": { "threshold": 0.55, "min_box_frac": 0.01 } },
   "target_class_names": ["sack", "half-sack"],
   "custom_model_path": null }
 ```
 
+`class_params` is `null`/omitted when no cell is filled — the job then runs the unchanged
+global path (REQ-181).
+
 #### 4.4.2 Mass Auto-Annotate modal
 
 Same anatomy, wider (1040 px), with three differences:
@@ -501,6 +511,10 @@ Same anatomy, wider (1040 px), with three differences:
   must not abort the rest; the backend GPU lock serialises the real work anyway. Progress is
   reported as `done/total (failed)` and the result message is `Queued N auto-annotation job(s)`.
 
+The per-class overrides table (REQ-181) appears here too, below the sliders, and its
+`class_params` ride along on every preview and start request — one set of overrides applies
+to all selected batches.
+
 ---
 
 ### 4.5 Review — `#/projects/{id}/review?batch={id}` or `#/batches/{id}`
@@ -1085,7 +1099,11 @@ network.
 - **Classes** — one row per class: a chip (swatch + name + its `1`-based hotkey) that sets the
   active class *and* reclasses the selected shape, plus a trash button that clears **every shape
   of that class across the whole batch** (`DELETE /batches/{id}/classes/{classId}/annotations`,
-  confirmed).
+  confirmed). Next to it sits an `×` button that clears that class **on the current frame only**
+  (REQ-180): one click, **no confirmation dialog**, it posts just this frame's shape ids of
+  that class to `POST /annotations/bulk-delete`, the row's count and the frame's badge update
+  optimistically with rollback on failure, and every other frame is untouched. The batch-wide
+  trash keeps its confirmation — that one is unrecoverable across the whole batch (REQ-046).
 - **Shapes on this frame (N)** — one row per annotation: swatch, class name, and either the
   score to 2 dp (auto) or the word `manual`; click selects, trash deletes. When empty:
   *"None on this frame — drag to draw a shape."* plus, if the batch has shapes elsewhere, a

frontend/src/components/AutoAnnotateModal.jsx

@@ -3,19 +3,8 @@ import { api } from '../api'
 import ExemplarCanvas from './ExemplarCanvas'
 import { PreviewShapes } from './PreviewShapes'
 import ClassPromptPanel from './ClassPromptPanel'
-
-function useDebounce(value, delay) {
-  const [debouncedValue, setDebouncedValue] = useState(value)
-  useEffect(() => {
-    const handler = setTimeout(() => {
-      setDebouncedValue(value)
-    }, delay)
-    return () => {
-      clearTimeout(handler)
-    }
-  }, [value, delay])
-  return debouncedValue
-}
+import ClassParamsTable, { buildClassParams } from './ClassParamsTable'
+import useDebounce from '../hooks/useDebounce'
 
 export default function AutoAnnotateModal({
   batch,
@@ -29,6 +18,7 @@ export default function AutoAnnotateModal({
   const [threshold, setThreshold] = useState(0.35)
   const [iouThreshold, setIouThreshold] = useState(0.0)
   const [minBoxFrac, setMinBoxFrac] = useState(0.0)
+  const [classParams, setClassParams] = useState({})
 
   
   // Class selection state
@@ -97,6 +87,7 @@ export default function AutoAnnotateModal({
       iou_threshold: iouThreshold,
       min_box_frac: minBoxFrac,
       target_class_names: selectedClasses,
+      class_params: buildClassParams(classParams),
       custom_model_path: customModelStagedPath,
       exemplars: engine === 'sam3' ? exemplars : [],
       exemplar_class_name: engine === 'sam3' ? activeClassName : null
@@ -195,6 +186,7 @@ export default function AutoAnnotateModal({
         iou_threshold: iouThreshold,
         min_box_frac: minBoxFrac,
         target_class_names: selectedClasses,
+        class_params: buildClassParams(classParams),
         custom_model_path: customModelStagedPath
       })
       onSuccess()
@@ -363,6 +355,13 @@ export default function AutoAnnotateModal({
             />
           </div>
 
+          <div style={{ marginBottom: 12 }}>
+            <span className="hint" style={{ fontSize: '0.8rem' }}>Per-class overrides (empty = global):</span>
+            <div style={{ marginTop: 4 }}>
+              <ClassParamsTable classNames={selectedClasses} globals={{ threshold, iou_threshold: iouThreshold, min_box_frac: minBoxFrac }} value={classParams} onChange={setClassParams} />
+            </div>
+          </div>
+
           <ClassPromptPanel
             classes={project.classes}
             selectedClasses={selectedClasses}

frontend/src/components/ClassParamsTable.jsx

@@ -0,0 +1,68 @@
+import React from 'react'
+
+// Per-class overrides of the job's globals (REQ-181). Empty input = inherit.
+const KEYS = [
+  { key: 'threshold', label: 'Conf', step: 0.05, color: '#38bdf8' },
+  { key: 'iou_threshold', label: 'IoU', step: 0.05, color: '#c084fc' },
+  { key: 'min_box_frac', label: 'MinBox', step: 0.005, color: '#34d399' },
+]
+
+export function buildClassParams(classParams) {
+  const out = {}
+  for (const [name, values] of Object.entries(classParams || {})) {
+    const entry = {}
+    for (const { key } of KEYS) {
+      const raw = values[key]
+      if (raw === '' || raw == null) continue
+      const num = parseFloat(raw)
+      if (Number.isFinite(num)) entry[key] = num
+    }
+    if (Object.keys(entry).length > 0) out[name] = entry
+  }
+  return Object.keys(out).length > 0 ? out : undefined
+}
+
+export default function ClassParamsTable({ classNames, globals, value, onChange }) {
+  const set = (name, key, raw) =>
+    onChange({ ...value, [name]: { ...(value[name] || {}), [key]: raw } })
+  if (!classNames.length) return null
+  return (
+    <table style={{ width: '100%', borderCollapse: 'collapse' }}>
+      <thead>
+        <tr>
+          <th className="hint" style={{ fontWeight: 400, fontSize: '0.74rem', textAlign: 'left', padding: '2px 4px' }}>Class</th>
+          {KEYS.map(k => (
+            <th key={k.key} className="hint" style={{ fontWeight: 400, fontSize: '0.74rem', color: k.color, padding: '2px 4px' }}>{k.label}</th>
+          ))}
+        </tr>
+      </thead>
+      <tbody>
+        {classNames.map(name => (
+          <tr key={name}>
+            <td className="mono" style={{ fontSize: '0.76rem', padding: '2px 4px' }}>{name}</td>
+            {KEYS.map(({ key, label, step, color }) => (
+              <td key={key} style={{ padding: '2px 4px' }}>
+                <input
+                  type="number"
+                  value={value[name]?.[key] ?? ''}
+                  min={0}
+                  max={1}
+                  step={step}
+                  placeholder={String(globals[key])}
+                  title={`Empty = use the global value (${globals[key]})`}
+                  aria-label={`${name} ${label} override, empty uses global ${globals[key]}`}
+                  onChange={(e) => set(name, key, e.target.value)}
+                  style={{
+                    width: 64, fontSize: '0.76rem', background: '#09090b', color,
+                    border: '1px solid #3f3f46', borderRadius: 4, padding: '2px 4px',
+                    cursor: 'text',
+                  }}
+                />
+              </td>
+            ))}
+          </tr>
+        ))}
+      </tbody>
+    </table>
+  )
+}

frontend/src/components/MassAutoAnnotateModal.jsx

@@ -1,6 +1,7 @@
 import React, { useEffect, useState } from 'react'
 import { api } from '../api'
 import { PreviewShapes } from './PreviewShapes'
+import ClassParamsTable, { buildClassParams } from './ClassParamsTable'
 
 const ENGINES = [
   { id: 'sam3', label: 'SAM3' },
@@ -17,6 +18,7 @@ export default function MassAutoAnnotateModal({ batches, project, onClose, onSuc
   const [threshold, setThreshold] = useState(0.35)
   const [iouThreshold, setIouThreshold] = useState(0.0)
   const [minBoxFrac, setMinBoxFrac] = useState(0.0)
+  const [classParams, setClassParams] = useState({})
   const [append, setAppend] = useState(true)
 
   const available = engine === 'custom' ? customClasses : project.classes.map(c => c.name)
@@ -81,6 +83,7 @@ export default function MassAutoAnnotateModal({ batches, project, onClose, onSuc
       iou_threshold: iouThreshold,
       min_box_frac: minBoxFrac,
       target_class_names: selectedClasses,
+      class_params: buildClassParams(classParams),
       custom_model_path: customPath,
     }).then(res => setPreviewShapes(res.shapes || []))
       .catch(exc => { setError(exc.message); setPreviewShapes([]) })
@@ -114,6 +117,7 @@ export default function MassAutoAnnotateModal({ batches, project, onClose, onSuc
           iou_threshold: iouThreshold,
           min_box_frac: minBoxFrac,
           target_class_names: selectedClasses,
+          class_params: buildClassParams(classParams),
           custom_model_path: customPath,
         })
       } catch {
@@ -243,6 +247,13 @@ export default function MassAutoAnnotateModal({ batches, project, onClose, onSuc
           <Slider label="Min Box Size (Fraction of Frame)" value={minBoxFrac} min={0} max={0.5} step={0.005}
                   color="#34d399" onChange={setMinBoxFrac} format={v => `${(v * 100).toFixed(1)}%`} />
 
+          <div style={{ marginBottom: 12 }}>
+            <span className="hint" style={{ fontSize: '0.8rem' }}>Per-class overrides (empty = global):</span>
+            <div style={{ marginTop: 4 }}>
+              <ClassParamsTable classNames={selectedClasses} globals={{ threshold, iou_threshold: iouThreshold, min_box_frac: minBoxFrac }} value={classParams} onChange={setClassParams} />
+            </div>
+          </div>
+
           <label className="row" style={{ gap: 6, alignItems: 'center', marginBottom: 12, cursor: 'pointer' }}>
             <input type="checkbox" checked={append} onChange={(e) => setAppend(e.target.checked)} style={{ cursor: 'pointer' }} />
             <span className="hint" style={{ fontSize: '0.8rem' }}>Append to existing annotations</span>

frontend/src/components/ReviewSidebar.jsx

@@ -1,6 +1,6 @@
 import React from 'react'
 import { classColor } from '../api'
-import { TrashIcon } from './Icons'
+import { TrashIcon, XIcon } from './Icons'
 import ShortcutsPanel from './ShortcutsPanel'
 
 export default function ReviewSidebar({
@@ -9,6 +9,7 @@ export default function ReviewSidebar({
   activeClass,
   reclass,
   clearClassInBatch,
+  clearClassInFrame,
   annotations,
   selectedId,
   setSelectedId,
@@ -23,7 +24,9 @@ export default function ReviewSidebar({
       <div className="panel side-panel">
         <h2>Classes</h2>
         <div className="class-list">
-          {classesList.map((item) => (
+          {classesList.map((item) => {
+            const onFrame = annotations.filter((row) => row.class_id === item.class_id).length
+            return (
             <div className="class-row" key={item.class_id} style={{ display: 'flex', alignItems: 'center', gap: 6 }}>
               <button
                 className={`class-chip ${item.class_id === activeClass ? 'active' : ''}`}
@@ -34,6 +37,15 @@ export default function ReviewSidebar({
                 {item.name}
                 <span className="faint mono">{item.class_id + 1}</span>
               </button>
+              <button
+                className="btn btn-ghost"
+                style={{ padding: '4px 6px' }}
+                title={`Clear ${onFrame} "${item.name}" shape${onFrame === 1 ? '' : 's'} on this frame`}
+                disabled={onFrame === 0}
+                onClick={() => clearClassInFrame(item)}
+              >
+                <XIcon size={12} />
+              </button>
               <button
                 className="btn btn-ghost"
                 style={{ padding: '4px 6px' }}
@@ -43,7 +55,8 @@ export default function ReviewSidebar({
                 <TrashIcon size={12} />
               </button>
             </div>
-          ))}
+            )
+          })}
         </div>
       </div>
 

frontend/src/hooks/useDebounce.js

@@ -0,0 +1,14 @@
+import { useState, useEffect } from 'react'
+
+export default function useDebounce(value, delay) {
+  const [debouncedValue, setDebouncedValue] = useState(value)
+  useEffect(() => {
+    const handler = setTimeout(() => {
+      setDebouncedValue(value)
+    }, delay)
+    return () => {
+      clearTimeout(handler)
+    }
+  }, [value, delay])
+  return debouncedValue
+}

frontend/src/pages/ReviewPage.jsx

@@ -392,6 +392,26 @@ export default function ReviewPage({ batchId: rawBatchId, projectId, onProject }
     } catch (exc) { setError(exc.message) }
   }
 
+  async function clearClassInFrame(item) {
+    const doomed = annotations.filter((row) => row.class_id === item.class_id)
+    if (!doomed.length || !frame) return
+    const ids = new Set(doomed.map((row) => row.id))
+    const previous = annotations
+    const previousCount = frame.annotation_count ?? 0
+    setAnnotations((rows) => rows.filter((row) => !ids.has(row.id)))
+    setMarkedIds((rows) => rows.filter((id) => !ids.has(id)))
+    patchFrameLocally(frame.id, { annotation_count: Math.max(0, previousCount - doomed.length) })
+    try {
+      await api.bulkDeleteAnnotations(doomed.map((row) => row.id))
+      setBatch(await api.getBatch(batchId))
+    } catch (exc) {
+      setAnnotations(previous)
+      setMarkedIds((rows) => [...rows, ...doomed.filter((row) => !rows.includes(row.id)).map((row) => row.id)])
+      patchFrameLocally(frame.id, { annotation_count: previousCount })
+      setError(exc.message)
+    }
+  }
+
   return (
     <>
       <div className="page-head">
@@ -602,6 +622,7 @@ export default function ReviewPage({ batchId: rawBatchId, projectId, onProject }
           activeClass={activeClass}
           reclass={reclass}
           clearClassInBatch={clearClassInBatch}
+          clearClassInFrame={clearClassInFrame}
           annotations={annotations}
           selectedId={selectedId}
           setSelectedId={setSelectedId}