update

python_server/app/routes/__init__.py

@@ -0,0 +1 @@
+"""HTTP 路由（按领域拆分）。"""

python_server/app/routes/animate.py

@@ -0,0 +1,135 @@
+from __future__ import annotations
+
+import datetime
+import shutil
+from typing import Any, Dict
+
+from fastapi import APIRouter
+from PIL import Image
+
+from ..deps import OUTPUT_DIR
+from ..schemas import AnimateRequest, CharacterAnimateRequest
+from ..services.animation_frames import (
+    align_size_to_input,
+    composite_rgba_on_background,
+    list_png_frames,
+    parse_background_color,
+    remove_background_by_color,
+)
+from ..services.image_io import b64_to_pil_image, pil_image_to_png_b64
+from ..services.predictors import get_animation_predictor
+
+router = APIRouter(tags=["animation"])
+
+
+@router.post("/animate")
+def animate(req: AnimateRequest) -> Dict[str, Any]:
+    try:
+        model_name = req.model_name or "animatediff"
+        predictor = get_animation_predictor(model_name)
+
+        result_path = predictor(
+            prompt=req.prompt,
+            negative_prompt=req.negative_prompt or "",
+            num_inference_steps=req.num_inference_steps,
+            guidance_scale=req.guidance_scale,
+            width=req.width,
+            height=req.height,
+            video_length=req.video_length,
+            seed=req.seed,
+        )
+
+        out_dir = OUTPUT_DIR / "animation"
+        out_dir.mkdir(parents=True, exist_ok=True)
+        ts = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+        out_path = out_dir / f"{model_name}_{ts}.gif"
+        shutil.copy2(result_path, out_path)
+
+        return {"success": True, "output_path": str(out_path)}
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+@router.post("/animate/character_sequence")
+def animate_character_sequence(req: CharacterAnimateRequest) -> Dict[str, Any]:
+    """
+    输入透明背景角色 PNG，生成 30fps 角色动画 PNG 序列。
+
+    流程：
+    - 将角色合成到指定纯色背景，作为 AnimateDiff 控制图
+    - 以与输入相近、且对齐到 8 的分辨率生成 PNG 序列
+    - 对每帧按背景色做色键剔除，返回透明背景 PNG 序列给前端
+    """
+    try:
+        model_name = req.model_name or "animatediff"
+        background_rgb = parse_background_color(req.background_color)
+
+        character = b64_to_pil_image(req.image_b64).convert("RGBA")
+        orig_w, orig_h = character.size
+        run_w, run_h = align_size_to_input(orig_w, orig_h, req.max_side)
+
+        predictor = get_animation_predictor(model_name)
+
+        out_dir = OUTPUT_DIR / "animation" / "character_sequence"
+        out_dir.mkdir(parents=True, exist_ok=True)
+        ts = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+        seq_dir = out_dir / f"{model_name}_{ts}"
+        seq_dir.mkdir(parents=True, exist_ok=True)
+
+        control_path = seq_dir / "control_image.png"
+        control_image = composite_rgba_on_background(
+            character,
+            background_rgb,
+            size=(run_w, run_h),
+        )
+        control_image.save(control_path)
+
+        frame_dir = predictor(
+            prompt=req.prompt,
+            negative_prompt=req.negative_prompt or "",
+            num_inference_steps=req.num_inference_steps,
+            guidance_scale=req.guidance_scale,
+            width=run_w,
+            height=run_h,
+            video_length=req.video_length,
+            seed=req.seed,
+            control_image_path=str(control_path),
+            output_format="png_sequence",
+        )
+
+        frame_paths = list_png_frames(frame_dir)
+        if not frame_paths:
+            raise FileNotFoundError(f"未找到 AnimateDiff 输出 PNG 序列: {frame_dir}")
+
+        frames: list[dict[str, Any]] = []
+        processed_dir = seq_dir / "frames"
+        processed_dir.mkdir(parents=True, exist_ok=True)
+        for idx, frame_path in enumerate(frame_paths):
+            frame = remove_background_by_color(
+                Image.open(frame_path),
+                background_rgb,
+                tolerance=req.background_tolerance,
+                output_size=(orig_w, orig_h),
+            )
+            out_path = processed_dir / f"{idx:04d}.png"
+            frame.save(out_path)
+            frames.append(
+                {
+                    "index": idx,
+                    "image_b64": pil_image_to_png_b64(frame),
+                    "path": str(out_path),
+                }
+            )
+
+        return {
+            "success": True,
+            "fps": 30,
+            "frame_count": len(frames),
+            "width": orig_w,
+            "height": orig_h,
+            "background_color": req.background_color,
+            "output_dir": str(processed_dir),
+            "frames": frames,
+        }
+    except Exception as e:
+        return {"success": False, "error": str(e), "frames": []}

python_server/app/routes/depth.py

@@ -0,0 +1,32 @@
+from __future__ import annotations
+
+import numpy as np
+from fastapi import APIRouter, HTTPException
+from fastapi.responses import Response
+
+from ..schemas import DepthRequest
+from ..services.depth_io import depth_to_png16_bytes
+from ..services.image_io import b64_to_pil_image
+from ..services.predictors import get_depth_predictor
+
+router = APIRouter(tags=["depth"])
+
+
+@router.post("/depth")
+def depth(req: DepthRequest):
+    """
+    计算深度并直接返回二进制 PNG（16-bit 灰度）。
+
+    约束：
+    - 前端不传/不选模型；模型选择写死在后端 config.py
+    - 成功：HTTP 200 + Content-Type: image/png
+    - 失败：HTTP 500，detail 为错误信息
+    """
+    try:
+        pil = b64_to_pil_image(req.image_b64).convert("RGB")
+        predictor = get_depth_predictor()
+        depth_arr = predictor(pil)  # type: ignore[misc]
+        png_bytes = depth_to_png16_bytes(np.asarray(depth_arr))
+        return Response(content=png_bytes, media_type="image/png")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e)) from e

python_server/app/routes/inpaint.py

@@ -0,0 +1,47 @@
+from __future__ import annotations
+
+from typing import Any, Dict
+
+from fastapi import APIRouter
+
+from ..deps import OUTPUT_DIR
+from ..schemas import InpaintRequest
+from ..services.image_io import b64_to_pil_image, default_half_mask, pil_image_to_png_b64
+from ..services.predictors import get_inpaint_predictor
+
+router = APIRouter(tags=["inpaint"])
+
+
+@router.post("/inpaint")
+def inpaint(req: InpaintRequest) -> Dict[str, Any]:
+    try:
+        model_name = req.model_name or "sdxl_inpaint"
+        pil = b64_to_pil_image(req.image_b64).convert("RGB")
+
+        if req.mask_b64:
+            mask = b64_to_pil_image(req.mask_b64).convert("L")
+        else:
+            mask = default_half_mask(pil)
+
+        predictor = get_inpaint_predictor(model_name)
+        out = predictor(
+            pil,
+            mask,
+            req.prompt or "",
+            req.negative_prompt or "",
+            strength=req.strength,
+            max_side=req.max_side,
+        )
+
+        out_dir = OUTPUT_DIR / "inpaint"
+        out_dir.mkdir(parents=True, exist_ok=True)
+        out_path = out_dir / f"{model_name}_inpaint.png"
+        out.save(out_path)
+
+        return {
+            "success": True,
+            "output_path": str(out_path),
+            "output_image_b64": pil_image_to_png_b64(out),
+        }
+    except Exception as e:
+        return {"success": False, "error": str(e)}

python_server/app/routes/meta.py

@@ -0,0 +1,93 @@
+from __future__ import annotations
+
+from typing import Any, Dict
+
+from fastapi import APIRouter
+
+router = APIRouter(tags=["meta"])
+
+
+@router.get("/models")
+def get_models() -> Dict[str, Any]:
+    """
+    返回一个兼容 Qt 前端的 schema：
+    {
+      "models": {
+        "depth":   { "key": { "name": "..."} ... },
+        "segment": { ... },
+        "inpaint": { "key": { "name": "...", "params": [...] } ... }
+      }
+    }
+    """
+    return {
+        "models": {
+            "depth": {
+                "midas": {"name": "MiDaS (default)"},
+                "zoedepth_n": {"name": "ZoeDepth (ZoeD_N)"},
+                "zoedepth_k": {"name": "ZoeDepth (ZoeD_K)"},
+                "zoedepth_nk": {"name": "ZoeDepth (ZoeD_NK)"},
+                "depth_anything_v2_vits": {"name": "Depth Anything V2 (vits)"},
+                "depth_anything_v2_vitb": {"name": "Depth Anything V2 (vitb)"},
+                "depth_anything_v2_vitl": {"name": "Depth Anything V2 (vitl)"},
+                "depth_anything_v2_vitg": {"name": "Depth Anything V2 (vitg)"},
+                "dpt_large": {"name": "DPT (large)"},
+                "dpt_hybrid": {"name": "DPT (hybrid)"},
+                "midas_dpt_beit_large_512": {"name": "MiDaS (dpt_beit_large_512)"},
+                "midas_dpt_swin2_large_384": {"name": "MiDaS (dpt_swin2_large_384)"},
+                "midas_dpt_swin2_tiny_256": {"name": "MiDaS (dpt_swin2_tiny_256)"},
+                "midas_dpt_levit_224": {"name": "MiDaS (dpt_levit_224)"},
+            },
+            "segment": {
+                "sam": {"name": "SAM (vit_h)"},
+                "mask2former_debug": {"name": "SAM (compat mask2former_debug)"},
+                "mask2former": {"name": "Mask2Former (not implemented)"},
+            },
+            "inpaint": {
+                "copy": {"name": "Copy (no-op)", "params": []},
+                "sdxl_inpaint": {
+                    "name": "SDXL Inpaint",
+                    "params": [
+                        {"id": "prompt", "label": "提示词", "optional": True},
+                    ],
+                },
+                "lama": {"name": "LaMa (Erase / Remove Object)", "params": []},
+                "controlnet": {
+                    "name": "ControlNet Inpaint (canny)",
+                    "params": [
+                        {"id": "prompt", "label": "提示词", "optional": True},
+                    ],
+                },
+            },
+            "animation": {
+                "animatediff": {
+                    "name": "AnimateDiff (Text-to-Video)",
+                    "params": [
+                        {"id": "prompt", "label": "提示词", "optional": False},
+                        {"id": "negative_prompt", "label": "负向提示词", "optional": True},
+                        {"id": "num_inference_steps", "label": "采样步数", "optional": True},
+                        {"id": "guidance_scale", "label": "CFG Scale", "optional": True},
+                        {"id": "width", "label": "宽度", "optional": True},
+                        {"id": "height", "label": "高度", "optional": True},
+                        {"id": "video_length", "label": "帧数", "optional": True},
+                        {"id": "seed", "label": "随机种子", "optional": True},
+                    ],
+                },
+                "character_sequence": {
+                    "name": "Character PNG Sequence (30 FPS)",
+                    "endpoint": "/animate/character_sequence",
+                    "params": [
+                        {"id": "image_b64", "label": "透明背景角色 PNG", "optional": False},
+                        {"id": "prompt", "label": "动画提示词", "optional": False},
+                        {"id": "background_color", "label": "纯色背景", "optional": True},
+                        {"id": "video_length", "label": "帧数", "optional": True},
+                        {"id": "background_tolerance", "label": "背景剔除阈值", "optional": True},
+                    ],
+                },
+            },
+        }
+    }
+
+
+@router.get("/health")
+def health() -> Dict[str, str]:
+    return {"status": "ok"}

python_server/app/routes/segment.py

@@ -0,0 +1,84 @@
+from __future__ import annotations
+
+from typing import Any, Dict
+
+import numpy as np
+from fastapi import APIRouter
+from PIL import Image
+
+from ..deps import OUTPUT_DIR
+from ..schemas import SamPromptSegmentRequest, SegmentRequest
+from ..services.image_io import b64_to_pil_image
+from ..services.predictors import get_seg_predictor
+from model.Seg.seg_loader import expand_mask, mask_to_contour_xy, run_sam_prompt
+
+router = APIRouter(tags=["segment"])
+
+
+@router.post("/segment")
+def segment(req: SegmentRequest) -> Dict[str, Any]:
+    try:
+        model_name = req.model_name or "sam"
+        pil = b64_to_pil_image(req.image_b64).convert("RGB")
+        rgb = np.array(pil, dtype=np.uint8)
+
+        predictor = get_seg_predictor(model_name)
+        label_map = predictor(rgb).astype(np.int32)
+
+        out_dir = OUTPUT_DIR / "segment"
+        out_dir.mkdir(parents=True, exist_ok=True)
+        out_path = out_dir / f"{model_name}_label.png"
+        Image.fromarray(np.clip(label_map, 0, 255).astype(np.uint8), mode="L").save(out_path)
+
+        return {"success": True, "label_path": str(out_path)}
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+@router.post("/segment/sam_prompt")
+def segment_sam_prompt(req: SamPromptSegmentRequest) -> Dict[str, Any]:
+    """
+    交互式 SAM：裁剪图 + 点/框提示，返回掩膜外轮廓点列（裁剪像素坐标）。
+    """
+    try:
+        pil = b64_to_pil_image(req.image_b64).convert("RGB")
+        rgb = np.array(pil, dtype=np.uint8)
+        h, w = rgb.shape[0], rgb.shape[1]
+
+        if req.overlay_b64:
+            ov = b64_to_pil_image(req.overlay_b64)
+            if ov.size != (w, h):
+                return {
+                    "success": False,
+                    "error": f"overlay 尺寸 {ov.size} 与 image {w}x{h} 不一致",
+                    "contour": [],
+                }
+
+        if len(req.point_coords) != len(req.point_labels):
+            return {
+                "success": False,
+                "error": "point_coords 与 point_labels 长度不一致",
+                "contour": [],
+            }
+        if len(req.point_coords) < 1:
+            return {"success": False, "error": "至少需要一个提示点", "contour": []}
+
+        pc = np.array(req.point_coords, dtype=np.float32)
+        if pc.ndim != 2 or pc.shape[1] != 2:
+            return {"success": False, "error": "point_coords 每项须为 [x,y]", "contour": []}
+
+        pl = np.array(req.point_labels, dtype=np.int64)
+        box = np.array(req.box_xyxy, dtype=np.float32)
+
+        mask = run_sam_prompt(rgb, pc, pl, box_xyxy=box)
+        if not np.any(mask):
+            return {"success": False, "error": "SAM 未产生有效掩膜", "contour": []}
+
+        mask = expand_mask(mask, int(req.expand_px))
+        contour = mask_to_contour_xy(mask, epsilon_px=2.0)
+        if len(contour) < 3:
+            return {"success": False, "error": "轮廓点数不足", "contour": []}
+
+        return {"success": True, "contour": contour, "error": None}
+    except Exception as e:
+        return {"success": False, "error": str(e), "contour": []}