Upload streamlit_app.py

streamlit_app.py

@@ -2,7 +2,7 @@
 # Talks to backend FastAPI: /health, /constraints/upload, /constraints/reset,
 # /detect, /plan, /plan_report, /analyze_xyz
 
-import os, io, uuid, base64, requests, math
+import os, io, uuid, base64, requests
 from typing import List, Tuple, Optional
 from PIL import Image, ImageDraw, ImageFont, ImageOps
 import streamlit as st
@@ -30,11 +30,10 @@ _defaults = {
     "API_BASE": DEFAULT_API_BASE,
     "size_preset": "1280×720",
     "jpeg_q": 90,
-    "keep_aspect": True,
     "aspect_mode": "16:9 crop",  # "Original", "16:9 crop", "4:3 crop", "1:1 crop"
-    "primary_labels": ["plate"],
-    "extra_labels": "food, burger",
-    "box_pick": "Center-most",  # "Highest score", "Largest area", "Center-most"
+    "primary_labels": ["plate", "food"],
+    "extra_labels": "",
+    "box_pick": "Highest score",  # "Highest score", "Largest area", "Center-most"
 }
 for k, v in _defaults.items():
     st.session_state.setdefault(k, v)
@@ -73,26 +72,21 @@ def _center_crop_aspect(img: Image.Image, target_ratio: float) -> Image.Image:
     if abs(cur_ratio - target_ratio) < 1e-3:
         return img
     if cur_ratio > target_ratio:
-        # too wide -> crop left/right
         new_w = int(H * target_ratio)
         x0 = (W - new_w) // 2
         return img.crop((x0, 0, x0 + new_w, H))
     else:
-        # too tall -> crop top/bottom
         new_h = int(W / target_ratio)
         y0 = (H - new_h) // 2
         return img.crop((0, y0, W, y0 + new_h))
 
 def _preprocess_image(file, target_w=1280, target_h=720,
-                      keep_aspect=True, aspect_mode="16:9 crop") -> Image.Image:
+                      aspect_mode="16:9 crop") -> Image.Image:
     img = Image.open(file).convert("RGB")
     if aspect_mode in _ASPECTS:
         img = _center_crop_aspect(img, _ASPECTS[aspect_mode])
-    # keep_aspect True => distort yok; True iken Image.resize zaten orana göre değil boyuta göre örnekler
-    # Bu yüzden önce crop ile oranı sabitliyoruz, sonra hedef boyuta resize ediyoruz.
-    img = img.resize((target_w, target_h), Image.Resampling.LANCZOS) if keep_aspect else ImageOps.fit(
-        img, (target_w, target_h), method=Image.Resampling.LANCZOS, centering=(0.5, 0.5)
-    )
+    # Esnetme yok — sadece yüksek kaliteli resize
+    img = img.resize((target_w, target_h), Image.Resampling.LANCZOS)
     return img
 
 def _b64_from_pil(img: Image.Image, jpeg_quality=90) -> str:
@@ -101,8 +95,8 @@ def _b64_from_pil(img: Image.Image, jpeg_quality=90) -> str:
     return base64.b64encode(buf.getvalue()).decode("utf-8")
 
 def _b64_from_file(file, target_w=1280, target_h=720, jpeg_quality=90,
-                   keep_aspect=True, aspect_mode="16:9 crop") -> Tuple[str, Image.Image]:
-    img = _preprocess_image(file, target_w, target_h, keep_aspect, aspect_mode)
+                   aspect_mode="16:9 crop") -> Tuple[str, Image.Image]:
+    img = _preprocess_image(file, target_w, target_h, aspect_mode)
     return _b64_from_pil(img, jpeg_quality), img
 
 # ================= Overlay drawing =================
@@ -146,20 +140,19 @@ def _draw_boxes(
 def _pick_primary_box(
     boxes: List[Tuple[float, float, float, float]],
     scores: List[float],
-    strategy: str = "Center-most",
+    strategy: str = "Highest score",
 ) -> Optional[int]:
     if not boxes:
         return None
     if strategy == "Largest area":
         areas = [(b[2]-b[0]) * (b[3]-b[1]) for b in boxes]
         return int(max(range(len(boxes)), key=lambda i: areas[i]))
-    if strategy == "Highest score":
-        return int(max(range(len(boxes)), key=lambda i: scores[i] if scores and i < len(scores) else -1e9))
-    # Center-most (default)
-    def center_dist(b):
-        cx = 0.5 * (b[0]+b[2]); cy = 0.5 * (b[1]+b[3])
-        return (cx-0.5)**2 + (cy-0.5)**2
-    return int(min(range(len(boxes)), key=lambda i: center_dist(boxes[i])))
+    if strategy == "Center-most":
+        def center_dist(b):
+            cx = 0.5 * (b[0]+b[2]); cy = 0.5 * (b[1]+b[3])
+            return (cx-0.5)**2 + (cy-0.5)**2
+        return int(min(range(len(boxes)), key=lambda i: center_dist(boxes[i])))
+    return int(max(range(len(boxes)), key=lambda i: scores[i] if scores and i < len(scores) else -1e9))
 
 def _detect_boxes_backend(image_b64: str, labels: List[str]) -> Optional[dict]:
     payload = {"image_b64": image_b64, "detect_query": ", ".join(labels)}
@@ -186,15 +179,12 @@ with st.sidebar:
         _success(f"Using {st.session_state['API_BASE']}")
         st.rerun()
     if colb2.button("Health"):
-        try:
-            r = _get("/health")
-            if r.status_code == 200:
-                _success("Backend reachable.")
-                st.json(r.json(), expanded=False)
-            else:
-                _err(f"{r.status_code}: {r.text[:300]}")
-        except Exception as e:
-            _err(f"Health error: {e}")
+        r = _get("/health")
+        if r.status_code == 200:
+            _success("Backend reachable.")
+            st.json(r.json(), expanded=False)
+        else:
+            _err(f"{r.status_code}: {r.text[:300]}")
 
     st.markdown("---")
     st.subheader("Rig Limits (export.txt)")
@@ -209,7 +199,11 @@ with st.sidebar:
                 r = _post_file("/constraints/upload", "file", export_file.read(), export_file.name)
                 if r.status_code == 200:
                     _success("Constraints loaded.")
-                    st.json(r.json(), expanded=False)
+                    data = r.json()
+                    st.json(data.get("parsed", {}), expanded=False)
+                    if "parser_log" in data:
+                        with st.expander("Parser log (constraints)"):
+                            st.json(data["parser_log"], expanded=False)
                 else:
                     _err(f"{r.status_code}: {r.text[:300]}")
     with colb:
@@ -247,7 +241,6 @@ with st.expander("Image optimization"):
         ["Original", "16:9 crop", "4:3 crop", "1:1 crop"],
         index=["Original", "16:9 crop", "4:3 crop", "1:1 crop"].index(st.session_state["aspect_mode"])
     )
-    st.session_state["keep_aspect"] = True  # Do not allow distortion in UI
 
 def _target_hw():
     preset = st.session_state.get("size_preset", "1280×720")
@@ -277,7 +270,7 @@ with col_text:
     all_suggestions = ["plate", "food", "burger", "person", "bottle", "product"]
     st.session_state["primary_labels"] = st.multiselect(
         "Primary labels (chips)",
-        options=sorted(set(all_suggestions + (st.session_state.get("primary_labels") or []))),
+        options=sorted(set(all_suggestions + st.session_state.get("primary_labels", []))),
         default=st.session_state["primary_labels"],
         key="primary_labels_widget",
     )
@@ -295,15 +288,7 @@ with col_text:
 def _gather_labels() -> List[str]:
     labels = list(st.session_state.get("primary_labels") or [])
     extra = [x.strip() for x in (st.session_state.get("extra_labels") or "").split(",") if x.strip()]
-    # Single label often more stable; still allow multi (cap to 12)
-    merged = [l for l in (labels + extra) if l]
-    # Deduplicate keeping order
-    seen = set()
-    out = []
-    for l in merged:
-        if l not in seen:
-            out.append(l); seen.add(l)
-    return out[:12] if out else ["subject"]
+    return [l for l in (labels + extra) if l][:12]
 
 # ---------- Detection preview card ----------
 det_card = st.container()
@@ -311,7 +296,6 @@ if img_file:
     tw, th = _target_hw()
     img_proc = _preprocess_image(
         img_file, tw, th,
-        keep_aspect=True,
         aspect_mode=st.session_state["aspect_mode"],
     )
     b64_img = _b64_from_pil(img_proc, int(st.session_state["jpeg_q"]))
@@ -321,22 +305,19 @@ if img_file:
 
         boxes_img = None
         try:
-            labels = _gather_labels()
+            labels = _gather_labels() or ["subject"]
             data = _detect_boxes_backend(b64_img, labels)
             if data and isinstance(data.get("boxes"), list) and len(data["boxes"]) > 0:
                 boxes = data["boxes"]
                 lbls = data.get("labels", ["object"] * len(boxes))
                 scrs = data.get("scores", [0.0] * len(boxes))
-                # Prefer backend's primary index if provided
-                primary_idx = data.get("primary_index", None)
-                if primary_idx is None or not (0 <= int(primary_idx) < len(boxes)):
-                    primary_idx = _pick_primary_box(boxes, scrs, st.session_state["box_pick"])
-                vis = _draw_boxes(img_proc, boxes, lbls, scrs, primary_idx=int(primary_idx) if primary_idx is not None else None)
+                pick = _pick_primary_box(boxes, scrs, st.session_state["box_pick"])
+                vis = _draw_boxes(img_proc, boxes, lbls, scrs, primary_idx=pick)
                 boxes_img = vis
         except Exception as e:
             _warn(f"Detection preview issue: {e}")
 
-        # Show at comfortable width, preserve aspect (no stretch)
+        # Esnetmeden, kullanışlı genişlikte göster (aspect otomatik korunur)
         max_w = 900
         if boxes_img is not None:
             st.image(boxes_img, caption="Detection overlay", width=max_w)
@@ -362,10 +343,9 @@ def _payload_prebuilt():
     img_b64, _img = _b64_from_file(
         img_file, tw, th,
         jpeg_quality=int(st.session_state["jpeg_q"]),
-        keep_aspect=True,
         aspect_mode=st.session_state["aspect_mode"],
     )
-    labels = _gather_labels()
+    labels = _gather_labels() or ["subject"]
     return {"instruction": instr.strip(), "image_b64": img_b64, "detect_query": ", ".join(labels)}
 
 # Generate-only
@@ -388,6 +368,9 @@ if btn_generate and _require_inputs():
             data = r.json()
             st.error("Violated constraints:"); st.write(data.get("violated", []))
             st.info("Suggestions:"); st.write(data.get("suggestions", []))
+            if "parser_log" in data:
+                with st.expander("Parser log (instruction)"):
+                    st.json(data["parser_log"], expanded=False)
         except Exception:
             _err(r.text[:600])
     else:
@@ -416,11 +399,9 @@ if btn_report and _require_inputs():
         if not ok:
             st.error("Violated constraints:"); st.write(data.get("violated", []))
             st.info("Suggestions:"); st.write(data.get("suggestions", []))
-        # diagnostics (if present)
-        diag = data.get("diag", {})
-        if diag:
-            with st.expander("Motion diagnostics"):
-                st.write(diag)
+        if "parser_log" in data:
+            with st.expander("Parser log (instruction)"):
+                st.json(data["parser_log"], expanded=False)
         st.caption("Use “Generate Path (.xyz)” to download the file.")
     else:
         _err(f"{r.status_code}: {r.text[:600]}")
@@ -445,4 +426,4 @@ st.caption(
     "Set env vars in this Space:\n"
     "• MotionPath_AI_API = https://<your-private-backend>.hf.space\n"
     "• MotionPath_AI_TOKEN = <same token as backend>"
-)
+)