Upload 5 files

Dockerfile

@@ -0,0 +1,20 @@
+# Dockerfile
+FROM python:3.10-slim
+
+ENV PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+# deps del sistema minimas
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential git && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY requirements.txt /app/
+RUN pip install --upgrade pip && pip install -r requirements.txt
+
+# copia todo (incluye tus .pt)
+COPY . /app
+
+# Hugging Face expone el puerto en $PORT; usamos ese valor
+CMD sh -c 'uvicorn app:app --host 0.0.0.0 --port ${PORT:-7860}'

requirements.txt

@@ -0,0 +1,6 @@
+fastapi
+uvicorn[standard]
+torch
+torchvision
+open-clip-torch
+Pillow

server1.py

@@ -0,0 +1,139 @@
+# app.py
+# comentarios sin tildes / sin enye
+
+import os, io
+from typing import Optional
+import torch
+from fastapi import FastAPI, File, UploadFile
+from fastapi.responses import JSONResponse
+from PIL import Image, UnidentifiedImageError
+import open_clip
+from torchvision import transforms as T
+
+# limites basicos
+torch.set_num_threads(1)
+os.environ["OMP_NUM_THREADS"] = "1"
+os.environ["MKL_NUM_THREADS"] = "1"
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+DTYPE  = torch.float16 if DEVICE == "cuda" else torch.float32
+
+# nombres de ficheros (en el mismo repo)
+MODEL_EMB_PATH = os.getenv("MODEL_EMB_PATH", "text_embeddings_modelos_h14.pt")
+VERS_EMB_PATH  = os.getenv("VERS_EMB_PATH",  "text_embeddings_h14.pt")
+
+app = FastAPI(title="CLIP H14 Vehicle API")
+
+# ============== modelo CLIP ==============
+clip_model, _, preprocess = open_clip.create_model_and_transforms(
+    "ViT-H-14", pretrained="laion2b_s32b_b79k"
+)
+clip_model = clip_model.to(device=DEVICE, dtype=DTYPE).eval()
+for p in clip_model.parameters():
+    p.requires_grad = False
+
+normalize = next(t for t in preprocess.transforms if isinstance(t, T.Normalize))
+transform = T.Compose([
+    T.Resize((224, 224), interpolation=T.InterpolationMode.BICUBIC),
+    T.ToTensor(),
+    T.Normalize(mean=normalize.mean, std=normalize.std),
+])
+
+# ============== embeddings ==============
+def _ensure_label_list(x):
+    if isinstance(x, (list, tuple)):
+        return list(x)
+    if hasattr(x, "tolist"):
+        return [str(s) for s in x.tolist()]
+    return [str(s) for s in x]
+
+def _load_embeddings(path: str):
+    ckpt = torch.load(path, map_location="cpu")
+    labels = _ensure_label_list(ckpt["labels"])
+    embeds = ckpt["embeddings"].to("cpu", dtype=torch.float16)
+    embeds = embeds / embeds.norm(dim=-1, keepdim=True)
+    return labels, embeds
+
+model_labels, model_embeddings   = _load_embeddings(MODEL_EMB_PATH)   # "Marca Modelo"
+version_labels, version_embeddings = _load_embeddings(VERS_EMB_PATH)  # "Marca Modelo Version"
+
+# ============== inferencia ==============
+@torch.inference_mode()
+def _encode_image(img_tensor: torch.Tensor) -> torch.Tensor:
+    if DEVICE == "cuda":
+        with torch.cuda.amp.autocast(dtype=DTYPE):
+            feats = clip_model.encode_image(img_tensor)
+    else:
+        feats = clip_model.encode_image(img_tensor)
+    return feats / feats.norm(dim=-1, keepdim=True)
+
+def _predict_top(text_feats_dev: torch.Tensor, text_labels: list[str], image_tensor: torch.Tensor, topk: int = 1):
+    img_f = _encode_image(image_tensor)
+    sim = (100.0 * img_f @ text_feats_dev.T).softmax(dim=-1)[0]
+    vals, idxs = torch.topk(sim, k=topk)
+    return [{"label": text_labels[i], "confidence": round(float(v)*100.0, 2)} for v, i in zip(vals, idxs)]
+
+def process_image_bytes(image_bytes: bytes):
+    if not image_bytes or len(image_bytes) < 128:
+        raise UnidentifiedImageError("imagen invalida")
+
+    img = Image.open(io.BytesIO(image_bytes))
+    if img.mode != "RGB":
+        img = img.convert("RGB")
+
+    img_tensor = transform(img).unsqueeze(0).to(device=DEVICE, dtype=DTYPE)
+
+    # paso 1: top-1 modelo
+    model_feats_dev = model_embeddings.to(DEVICE) if DEVICE == "cuda" else model_embeddings
+    top_model = _predict_top(model_feats_dev, model_labels, img_tensor, topk=1)[0]
+    modelo_full = top_model["label"]; conf_m = top_model["confidence"]
+
+    partes = modelo_full.split(" ", 1)
+    marca  = partes[0] if len(partes) >= 1 else ""
+    modelo = partes[1] if len(partes) == 2 else ""
+
+    # paso 2: filtrar versiones por prefijo
+    matches = [(lab, idx) for idx, lab in enumerate(version_labels) if lab.startswith(modelo_full)]
+    if not matches:
+        return {
+            "marca": marca.upper(), "modelo": modelo.title(),
+            "version": "", "confianza_modelo": conf_m, "confianza_version": 0.0
+        }
+
+    idxs = [i for _, i in matches]
+    labels_sub = [lab for lab, _ in matches]
+    embeds_sub = version_embeddings[idxs].to(DEVICE) if DEVICE == "cuda" else version_embeddings[idxs]
+
+    # paso 3: top-1 version
+    top_ver = _predict_top(embeds_sub, labels_sub, img_tensor, topk=1)[0]
+    raw = top_ver["label"]; conf_v = top_ver["confidence"]
+
+    prefix = modelo_full + " "
+    ver = raw[len(prefix):] if raw.startswith(prefix) else raw
+    ver = ver.split(" ")[0]
+    if conf_v < 25.0:
+        ver = "Version no identificada con suficiente confianza"
+
+    return {
+        "marca": marca.upper(),
+        "modelo": modelo.title(),
+        "version": ver.title() if ver else "",
+        "confianza_modelo": conf_m,
+        "confianza_version": conf_v
+    }
+
+# ============== endpoints ==============
+@app.get("/")
+def root():
+    return {"status": "ok", "device": DEVICE}
+
+@app.post("/predict")
+async def predict(front: UploadFile = File(...), back: Optional[UploadFile] = File(None)):
+    try:
+        front_bytes = await front.read()
+        if back is not None:
+            _ = await back.read()
+        result = process_image_bytes(front_bytes)
+        return JSONResponse(content={"code": 200, "data": result})
+    except Exception:
+        return JSONResponse(content={"code": 404, "data": {}}, status_code=200)

text_embeddings_h14.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d2840ca4c8c94783b39b166e8144f7fe084dc91d9aae38f1042c9732b385bf2
+size 2346749

text_embeddings_modelos_h14.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4f6be76db1d261daf5ca4e29e491cee866100101c81cb0829da9e2c7b4afd6c
+size 843829