Spaces:
Sleeping
Sleeping
File size: 8,645 Bytes
a3953cb | 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 | """
generate.py — POST /generate
Receives a job_id and mask JSON, determines the base image from the current
step's result_image, creates a new step, and kicks off LaMa inpainting as a
FastAPI BackgroundTask.
Step creation logic
-------------------
1. Read job.current_step (call it N).
2. Use step N's result_image as the base image for the new step.
3. New step number = N + 1.
4. If step N+1 already exists (user went back then edited again):
- Truncate the steps list to keep only 0..N.
- Delete the orphaned step folders from disk to avoid clutter.
This implements linear history: branching is not supported; new work after
"Go Back" discards the forward history, like undo/redo in most editors.
5. Copy the base image to steps/{N+1}/original.jpg so each step folder is
self-contained.
6. Write masks to steps/{N+1}/mask_hard.png and mask_soft.png.
7. Append a new StepRecord at step N+1 with status="processing".
8. Advance job.current_step to N+1.
9. Enqueue background inference.
Engine access pattern
---------------------
The LamaEngine is fetched from request.app.state while the Request is alive,
then passed as a plain argument to run_inpaint_job() so it has no dependency
on the Request object after the response is sent.
"""
from __future__ import annotations
import shutil
import traceback
from pathlib import Path
from typing import Literal
import cv2
import numpy as np
from fastapi import APIRouter, BackgroundTasks, HTTPException, Request
from pydantic import BaseModel
from app.config import STORAGE_DIR
from app.utils import job_folder_name
from app.jobs_store import (
get_job,
update_job,
add_step,
update_step,
truncate_steps_after,
)
from app.services.mask_service import build_masks, save_mask
from app.services.blend_service import blend, check_output_quality
from app.services.inpaint_engine import BaseInpaintEngine
from app.models.schemas import StepRecord
router = APIRouter()
# ---------------------------------------------------------------------------
# Request schema
# ---------------------------------------------------------------------------
class GenerateRequest(BaseModel):
job_id: str
mask_source: Literal["manual", "ocr", "mixed"] = "manual"
mask_json: dict # validated further inside mask_service.build_masks()
# ---------------------------------------------------------------------------
# Background worker
# ---------------------------------------------------------------------------
def run_inpaint_job(
job_id: str,
folder_name: str,
step_number: int,
image: np.ndarray,
hard_mask: np.ndarray,
soft_mask: np.ndarray,
engine: BaseInpaintEngine,
) -> None:
"""Run LaMa inpainting and blend, then update the step's status.
Called by FastAPI BackgroundTasks after the /generate response is sent.
Updates the StepRecord to "completed" or "failed" — never raises.
folder_name is the human-readable on-disk name (job_folder_name(job_id, created_at)),
passed explicitly so the worker never needs to re-read the job from disk.
"""
step_dir = STORAGE_DIR / "watermark" / folder_name / "steps" / str(step_number)
try:
# Step 1 — run LaMa inpainting
lama_output: np.ndarray = engine.inpaint(image, hard_mask)
# Step 2 — composite output onto original using feathered mask
result_image: np.ndarray = blend(image, lama_output, soft_mask)
# Step 3 — quality check before saving
passed, error_msg = check_output_quality(result_image, expected_shape=image.shape)
if not passed:
update_step(job_id, step_number, status="failed", error_message=error_msg)
return
# Step 4 — save result
result_path = step_dir / "result.jpg"
ok = cv2.imwrite(str(result_path), result_image)
if not ok:
update_step(
job_id, step_number,
status="failed",
error_message="cv2.imwrite failed writing result.jpg.",
)
return
# Step 5 — mark completed
update_step(
job_id, step_number,
status="completed",
result_image=str(result_path),
)
except Exception:
update_step(
job_id, step_number,
status="failed",
error_message=traceback.format_exc(),
)
# ---------------------------------------------------------------------------
# Route
# ---------------------------------------------------------------------------
@router.post("")
async def generate(
body: GenerateRequest,
background_tasks: BackgroundTasks,
request: Request,
) -> dict:
"""Validate mask JSON, create new step, enqueue background inpainting.
Returns immediately with {"job_id": ..., "step_number": ..., "status": "processing"}.
Inference has NOT started at the point this response is sent.
"""
job_id = body.job_id
# --- Confirm job exists --------------------------------------------------
job = get_job(job_id)
if job is None:
raise HTTPException(status_code=404, detail=f"Job '{job_id}' not found.")
# Compute the on-disk folder name once; reused for all path construction below.
# job.created_at is immutable after creation so this always yields the same string.
folder_name = job_folder_name(job.id, job.created_at)
# --- Locate current step and base image ----------------------------------
current_step_num = job.current_step
current_step = next(
(s for s in job.steps if s.step_number == current_step_num), None
)
if current_step is None or current_step.result_image is None:
raise HTTPException(
status_code=400,
detail=f"Step {current_step_num} has no result image to build from.",
)
base_path = Path(current_step.result_image)
if not base_path.exists():
raise HTTPException(
status_code=500,
detail=f"Base image missing from storage: {base_path}",
)
# --- Load base image -----------------------------------------------------
image: np.ndarray = cv2.imread(str(base_path))
if image is None:
raise HTTPException(
status_code=500,
detail=f"cv2.imread could not decode: {base_path}",
)
# --- Build masks ---------------------------------------------------------
try:
hard_mask, soft_mask = build_masks(body.mask_json)
except ValueError as exc:
raise HTTPException(status_code=422, detail=str(exc))
# --- Truncate any forward history (linear history: no branching) ---------
new_step_num = current_step_num + 1
if any(s.step_number >= new_step_num for s in job.steps):
removed = truncate_steps_after(job_id, keep_through=current_step_num)
for s in removed:
orphan = STORAGE_DIR / "watermark" / folder_name / "steps" / str(s["step_number"])
if orphan.exists():
shutil.rmtree(orphan)
# --- Create step folder and copy base image as this step's original ------
step_dir = STORAGE_DIR / "watermark" / folder_name / "steps" / str(new_step_num)
step_dir.mkdir(parents=True, exist_ok=True)
step_original = step_dir / "original.jpg"
shutil.copy2(str(base_path), str(step_original))
# --- Save masks ----------------------------------------------------------
hard_mask_path = step_dir / "mask_hard.png"
soft_mask_path = step_dir / "mask_soft.png"
save_mask(hard_mask, str(hard_mask_path))
save_mask(soft_mask, str(soft_mask_path))
# --- Append new step record (status=processing) --------------------------
new_step = StepRecord(
step_number=new_step_num,
original_image=str(step_original),
mask_image=str(hard_mask_path),
mask_source=body.mask_source,
status="processing",
)
add_step(job_id, new_step)
# --- Advance current_step ------------------------------------------------
update_job(job_id, current_step=new_step_num)
# --- Enqueue background inference ----------------------------------------
engine: BaseInpaintEngine = request.app.state.inpaint_engine
background_tasks.add_task(
run_inpaint_job,
job_id = job_id,
folder_name = folder_name,
step_number = new_step_num,
image = image,
hard_mask = hard_mask,
soft_mask = soft_mask,
engine = engine,
)
return {"job_id": job_id, "step_number": new_step_num, "status": "processing"}
|