File size: 16,317 Bytes
f9c869c 8adeaa1 f9c869c 8adeaa1 f9c869c d35edd7 9adcd23 f9c869c ac44754 69c768f e31116c d35edd7 8adeaa1 f9c869c 9adcd23 ac44754 9adcd23 f9c869c 8adeaa1 d35edd7 8adeaa1 d35edd7 8adeaa1 9adcd23 f9c869c 69c768f f9c869c c6c1400 f9c869c 69c768f 9adcd23 f9c869c e31116c f9c869c e31116c d35edd7 8adeaa1 d35edd7 8adeaa1 e31116c 8adeaa1 d35edd7 e31116c f9c869c 9adcd23 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 | """HTTP server for a GeoText-1652 custom Inference Endpoint container."""
from __future__ import annotations
import os
from contextlib import asynccontextmanager
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
import numpy as np
import torch
from fastapi import FastAPI, HTTPException
from hub_storage import persist_patch_embeddings
from infer import (
fetch_rgb_geotiff,
image_embedding_outputs,
load_model,
model_outputs,
text_embedding_outputs,
tile_image,
)
from pydantic import BaseModel, Field
class InferenceRequest(BaseModel):
image_url: str = Field(..., description="Public RGB GeoTIFF/COG, JPEG, or PNG URL")
queries: List[str] = Field(..., min_items=1, description="Text queries to rank")
include_embeddings: bool = Field(
False, description="Return 256-dimensional image and text vectors"
)
include_bboxes: bool = Field(False, description="Return text-conditioned normalized boxes")
include_patch_embeddings: bool = Field(
False,
description="Return normalized 256-dimensional spatial patch embeddings",
)
store_patch_embeddings: bool = Field(
False, description="Persist patch embeddings as Parquet in the configured Hub bucket"
)
output_prefix: Optional[str] = Field(
None, description="Optional Hub bucket key prefix for persisted patch embeddings"
)
tile_overlap: int = Field(64, ge=0, description="Source-pixel overlap between tiles")
tile_size: Optional[int] = Field(
None, ge=64, description="Optional source-pixel tile size for tiled inference"
)
return_tile_results: bool = Field(
False, description="Include per-tile scores and regions in addition to stitched results"
)
class TextEmbeddingRequest(BaseModel):
queries: List[str] = Field(..., min_items=1, description="Text strings to embed")
class ImageEmbeddingRequest(BaseModel):
image_url: str = Field(..., description="Public RGB GeoTIFF/COG, JPEG, or PNG URL")
include_global_embedding: bool = Field(
True, description="Return the normalized 256-dimensional image vector"
)
include_patch_embeddings: bool = Field(
False, description="Return normalized 256-dimensional spatial patch vectors"
)
store_patch_embeddings: bool = Field(
False, description="Persist patch embeddings as Parquet in HF_BUCKET"
)
output_prefix: Optional[str] = Field(
None, description="Optional Hugging Face dataset path prefix"
)
tile_overlap: int = Field(64, ge=0, description="Source-pixel overlap between tiles")
tile_size: Optional[int] = Field(
None, ge=64, description="Optional source-pixel tile size"
)
return_tile_results: bool = Field(
False, description="Include per-tile embeddings in addition to stitched output"
)
@asynccontextmanager
async def lifespan(app: FastAPI):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_dir = Path(os.environ.get("GEOTEXT_MODEL_DIR", "/repository"))
source_dir = Path(os.environ.get("GEOTEXT_SOURCE_DIR", "/app/GeoText-1652"))
model, tokenizer, config = load_model(source_dir, model_dir, device)
app.state.runtime = {
"config": config,
"device": device,
"model": model,
"tokenizer": tokenizer,
}
yield
app = FastAPI(title="GeoText-1652", lifespan=lifespan)
def _pixel_box(box: List[float], tile: Dict[str, Any]) -> List[float]:
"""Convert a tile-local normalized cx/cy/w/h box into source pixel xyxy."""
cx, cy, width, height = box
x1 = max(tile["column"], tile["column"] + (cx - width / 2) * tile["width"])
y1 = max(tile["row"], tile["row"] + (cy - height / 2) * tile["height"])
x2 = min(tile["column"] + tile["width"], tile["column"] + (cx + width / 2) * tile["width"])
y2 = min(tile["row"] + tile["height"], tile["row"] + (cy + height / 2) * tile["height"])
return [float(x1), float(y1), float(x2), float(y2)]
def _iou(left: List[float], right: List[float]) -> float:
x1 = max(left[0], right[0])
y1 = max(left[1], right[1])
x2 = min(left[2], right[2])
y2 = min(left[3], right[3])
intersection = max(0.0, x2 - x1) * max(0.0, y2 - y1)
if not intersection:
return 0.0
left_area = max(0.0, left[2] - left[0]) * max(0.0, left[3] - left[1])
right_area = max(0.0, right[2] - right[0]) * max(0.0, right[3] - right[1])
return intersection / max(left_area + right_area - intersection, 1e-8)
def _stitch_boxes(
tile_outputs: List[Tuple[Dict[str, Any], Dict[str, Any]]], image_width: int, image_height: int
) -> List[Dict[str, Any]]:
candidates = []
for tile, output in tile_outputs:
score_by_query = {item["query"]: item["similarity"] for item in output["ranked_queries"]}
for region in output.get("text_conditioned_boxes", []):
pixel_box = _pixel_box(region["box"], tile)
candidates.append(
{
"box": pixel_box,
"query": region["query"],
"score": score_by_query[region["query"]],
"tile_index": tile["index"],
}
)
stitched = []
for query in sorted({candidate["query"] for candidate in candidates}):
pending = sorted(
(candidate for candidate in candidates if candidate["query"] == query),
key=lambda candidate: candidate["score"],
reverse=True,
)
while pending:
selected = pending.pop(0)
stitched.append(
{
"query": query,
"score": selected["score"],
"source_pixel_xyxy": selected["box"],
"source_normalized_xyxy": [
selected["box"][0] / image_width,
selected["box"][1] / image_height,
selected["box"][2] / image_width,
selected["box"][3] / image_height,
],
"tile_index": selected["tile_index"],
}
)
pending = [
candidate for candidate in pending if _iou(selected["box"], candidate["box"]) < 0.5
]
return stitched
def _tiled_outputs(
runtime: Dict[str, Any], image, request: InferenceRequest
) -> Dict[str, Any]:
tiles = tile_image(image, request.tile_size, request.tile_overlap)
tile_outputs = []
for tile in tiles:
output = model_outputs(
runtime["model"], runtime["tokenizer"], runtime["config"], tile["image"],
request.queries,
runtime["device"],
request.include_embeddings,
request.include_bboxes,
request.include_patch_embeddings or request.store_patch_embeddings,
)
tile_outputs.append((tile, output))
best_by_query = {}
for tile, output in tile_outputs:
for item in output["ranked_queries"]:
candidate = dict(item, tile_index=tile["index"])
if (
item["query"] not in best_by_query
or item["similarity"] > best_by_query[item["query"]]["similarity"]
):
best_by_query[item["query"]] = candidate
result = {
"ranked_queries": sorted(
best_by_query.values(), key=lambda item: item["similarity"], reverse=True
),
"tiling": {
"tile_size": request.tile_size,
"tile_overlap": request.tile_overlap,
"tile_count": len(tiles),
"similarity_stitching": "maximum tile similarity per query",
},
}
if request.include_embeddings:
weights = np.asarray(
[tile["width"] * tile["height"] for tile, _ in tile_outputs], dtype=np.float32
)
vectors = np.asarray(
[output["image_embedding"] for _, output in tile_outputs], dtype=np.float32
)
image_embedding = np.average(vectors, axis=0, weights=weights)
image_embedding /= max(float(np.linalg.norm(image_embedding)), 1e-8)
result["image_embedding"] = image_embedding.tolist()
result["text_embeddings"] = tile_outputs[0][1]["text_embeddings"]
if request.include_bboxes:
result["stitched_text_conditioned_boxes"] = _stitch_boxes(
tile_outputs, image.width, image.height
)
if request.include_patch_embeddings or request.store_patch_embeddings:
patches = []
for tile, output in tile_outputs:
for patch in output["patch_embeddings"]["patches"]:
x1, y1, x2, y2 = patch["source_pixel_xyxy"]
patches.append(
{
**patch,
"tile_index": tile["index"],
"source_pixel_xyxy": [
x1 + tile["column"],
y1 + tile["row"],
x2 + tile["column"],
y2 + tile["row"],
],
}
)
result["patch_embeddings"] = {
"embedding_dimension": tile_outputs[0][1]["patch_embeddings"]["embedding_dimension"],
"grid": "per_tile",
"patches": patches,
}
if request.return_tile_results:
result["tile_results"] = [
{
"tile_index": tile["index"],
"source_pixel_window": [
tile["column"],
tile["row"],
tile["width"],
tile["height"],
],
**output,
}
for tile, output in tile_outputs
]
return result
def _image_outputs(
runtime: Dict[str, Any], image, request: ImageEmbeddingRequest
) -> Dict[str, Any]:
"""Run vision-only embedding inference, optionally over overlapping tiles."""
include_patches = request.include_patch_embeddings or request.store_patch_embeddings
if request.tile_size is None:
return image_embedding_outputs(
runtime["model"],
runtime["config"],
image,
runtime["device"],
include_global_embedding=request.include_global_embedding,
include_patch_embeddings=include_patches,
)
tiles = tile_image(image, request.tile_size, request.tile_overlap)
tile_outputs = []
for tile in tiles:
output = image_embedding_outputs(
runtime["model"],
runtime["config"],
tile["image"],
runtime["device"],
include_global_embedding=request.include_global_embedding,
include_patch_embeddings=include_patches,
)
tile_outputs.append((tile, output))
result = {
"embedding_dimension": tile_outputs[0][1]["embedding_dimension"],
"tiling": {
"tile_size": request.tile_size,
"tile_overlap": request.tile_overlap,
"tile_count": len(tiles),
"global_embedding_stitching": "area-weighted mean, then L2 normalization",
},
}
if request.include_global_embedding:
weights = np.asarray(
[tile["width"] * tile["height"] for tile, _ in tile_outputs], dtype=np.float32
)
vectors = np.asarray(
[output["image_embedding"] for _, output in tile_outputs], dtype=np.float32
)
embedding = np.average(vectors, axis=0, weights=weights)
embedding /= max(float(np.linalg.norm(embedding)), 1e-8)
result["image_embedding"] = embedding.tolist()
if include_patches:
patches = []
for tile, output in tile_outputs:
for patch in output["patch_embeddings"]["patches"]:
x1, y1, x2, y2 = patch["source_pixel_xyxy"]
patches.append(
{
**patch,
"tile_index": tile["index"],
"source_pixel_xyxy": [
x1 + tile["column"],
y1 + tile["row"],
x2 + tile["column"],
y2 + tile["row"],
],
}
)
result["patch_embeddings"] = {
"embedding_dimension": result["embedding_dimension"],
"grid": "per_tile",
"patches": patches,
}
if request.return_tile_results:
result["tile_results"] = [
{
"tile_index": tile["index"],
"source_pixel_window": [
tile["column"], tile["row"], tile["width"], tile["height"]
],
**output,
}
for tile, output in tile_outputs
]
return result
@app.get("/health")
def health() -> Dict[str, str]:
return {"status": "ok"}
@app.post("/")
@app.post("/infer")
def infer(request: InferenceRequest) -> Dict[str, Any]:
"""Backward-compatible combined image/text search endpoint."""
try:
image, image_metadata = fetch_rgb_geotiff(request.image_url)
runtime = app.state.runtime
result = {
"image_url": request.image_url,
"image_metadata": image_metadata,
"device": str(runtime["device"]),
}
include_patches = request.include_patch_embeddings or request.store_patch_embeddings
if request.tile_size is None:
result.update(
model_outputs(
runtime["model"],
runtime["tokenizer"],
runtime["config"],
image,
request.queries,
runtime["device"],
include_embeddings=request.include_embeddings,
include_bboxes=request.include_bboxes,
include_patch_embeddings=include_patches,
)
)
else:
result.update(_tiled_outputs(runtime, image, request))
if request.store_patch_embeddings:
result["patch_embeddings_storage"] = persist_patch_embeddings(
image_url=request.image_url,
patches=result["patch_embeddings"]["patches"],
output_prefix=request.output_prefix,
)
if not request.include_patch_embeddings:
result.pop("patch_embeddings")
return result
except Exception as error:
raise HTTPException(status_code=422, detail=str(error)) from error
@app.post("/embed/text")
def embed_text(request: TextEmbeddingRequest) -> Dict[str, Any]:
"""Return query vectors without fetching imagery or running the vision encoder."""
try:
runtime = app.state.runtime
result = text_embedding_outputs(
runtime["model"],
runtime["tokenizer"],
runtime["config"],
request.queries,
runtime["device"],
)
return {"device": str(runtime["device"]), **result}
except Exception as error:
raise HTTPException(status_code=422, detail=str(error)) from error
@app.post("/embed/image")
def embed_image(request: ImageEmbeddingRequest) -> Dict[str, Any]:
"""Return image or patch vectors without tokenizing text or running search."""
try:
image, image_metadata = fetch_rgb_geotiff(request.image_url)
runtime = app.state.runtime
result = {
"image_url": request.image_url,
"image_metadata": image_metadata,
"device": str(runtime["device"]),
}
result.update(_image_outputs(runtime, image, request))
if request.store_patch_embeddings:
result["patch_embeddings_storage"] = persist_patch_embeddings(
image_url=request.image_url,
patches=result["patch_embeddings"]["patches"],
output_prefix=request.output_prefix,
)
if not request.include_patch_embeddings:
result.pop("patch_embeddings")
return result
except Exception as error:
raise HTTPException(status_code=422, detail=str(error)) from error
|