fix(Aesthetic Scorer): Remove debug and nesting response

Author: danactive

app/api/admin/classify/route.ts

@@ -13,9 +13,6 @@ export async function POST(req: Request) {
   try {
     const utils = utilsFactory()
 
-    const url = new URL(req.url)
-    const debug = url.searchParams.get('debug') === 'true'
-
     const { path: relativePath } = await req.json()
 
     if (!relativePath) {
@@ -25,7 +22,7 @@ export async function POST(req: Request) {
     const fullPath = utils.safePublicPath(relativePath)
     const buffer = await fs.readFile(fullPath)
 
-    const classifyUrl = `http://localhost:${config.pythonPort}/classify?debug=${debug}`
+    const classifyUrl = `http://localhost:${config.pythonPort}/classify`
 
     const res = await fetch(classifyUrl, {
       method: 'POST',

app/api/admin/scores/route.ts

@@ -1,16 +1,13 @@
-import { NextRequest, NextResponse } from 'next/server'
+import { NextResponse } from 'next/server'
 import fs from 'node:fs/promises'
 
 import utilsFactory from '../../../../src/lib/utils'
 import config from '../../../../src/models/config'
 
-async function POST(req: Request) {
+export async function POST(req: Request) {
   try {
     const utils = utilsFactory()
 
-    const url = new URL(req.url)
-    const debug = url.searchParams.get('debug') === 'true'
-
     const { path: relativePath } = await req.json()
 
     if (!relativePath) {
@@ -20,7 +17,7 @@ async function POST(req: Request) {
     const fullPath = utils.safePublicPath(relativePath)
     const buffer = await fs.readFile(fullPath)
 
-    const classifyUrl = `http://localhost:${config.pythonPort}/scores?debug=${debug}`
+    const classifyUrl = `http://localhost:${config.pythonPort}/scores`
 
     const res = await fetch(classifyUrl, {
       method: 'POST',
@@ -42,18 +39,3 @@ async function POST(req: Request) {
     return NextResponse.json({ error: err.message || 'Unexpected error' }, { status: 500 })
   }
 }
-
-// Catch-all for unsupported methods
-function notSupported(req: NextRequest) {
-  return NextResponse.json(`Method ${req.method} Not Allowed`, { status: 405 })
-}
-
-export {
-  notSupported as GET,
-  POST,
-  notSupported as PUT,
-  notSupported as DELETE,
-  notSupported as PATCH,
-  notSupported as OPTIONS,
-  notSupported as HEAD,
-}

apps/api/aesthetic.py

@@ -1,10 +1,12 @@
+from fastapi import Request
 import torch
 import torch.nn as nn
-from PIL import Image
 import torchvision.transforms as T
 import open_clip
+from PIL import Image
 import logging
 from collections import OrderedDict
+import io
 
 # Set up logging once
 logging.basicConfig(level=logging.DEBUG)
@@ -69,21 +71,14 @@ def load_clip_model() -> tuple[torch.nn.Module, callable]:
 _clip_model, preprocess = load_clip_model()
 regression_head = load_aesthetic_head(HEAD_PATH)
 
-def score_aesthetic(pil_image: Image.Image) -> float:
-    model, _, preprocess = open_clip.create_model_and_transforms(
-        'ViT-L-14',
-        pretrained=None
-    )
-    model.eval()
-
-    logger.info("Load your checkpoint...")
-    logger.info("Load your regression head...")
-
-    with torch.no_grad():
-        image_tensor = preprocess(pil_image).unsqueeze(0)
-        image_features = model.encode_image(image_tensor)
-        image_features /= image_features.norm(dim=-1, keepdim=True)
-        score = regression_head(image_features).item()
+async def score_aesthetic(req: Request) -> float:
+  img_bytes = await req.body()
+  img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
 
-    return float(score)
+  with torch.no_grad():
+    image_tensor = preprocess(img).unsqueeze(0)
+    image_features = _clip_model.encode_image(image_tensor)
+    image_features /= image_features.norm(dim=-1, keepdim=True)
+    score = regression_head(image_features).item()
 
+  return float(score)

apps/api/classify.py

@@ -1,10 +1,9 @@
 from fastapi import Request
 import torch
 import timm
-import json
-from PIL import Image
 import torchvision.transforms as T
-import base64
+from PIL import Image
+import json
 import io
 import logging
 
@@ -49,12 +48,11 @@
   T.CenterCrop(input_size),
 ])
 
-async def classify_image(req: Request, debug: bool):
+async def classify_image(req: Request):
   img_bytes = await req.body()
   img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
 
   # Image preparation
-  resized_img = debug_transform(img)
   input_tensor = transform(img).unsqueeze(0)
 
   with torch.no_grad():
@@ -74,29 +72,4 @@ async def classify_image(req: Request, debug: bool):
   if not isinstance(results, list):
     results = [results]
 
-  if not debug:
-    return {"predictions": results}, {}
-
-  # Debug details
-  topk_debug = torch.topk(probs, k=50)
-  top_50 = [
-    {
-      "label": idx_to_label.get(str(idx.item()), f"Unknown ({idx.item()})"),
-      "score": float(score)
-    }
-    for idx, score in zip(topk_debug.indices[0], topk_debug.values[0])
-  ]
-
-  buffered = io.BytesIO()
-  resized_img.save(buffered, format="JPEG")
-  img_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
-
-  return {
-    "predictions": results,
-    "top_50": top_50,
-    "logits_shape": list(logits.shape),
-    "max_prob": float(probs.max()),
-    "min_prob": float(probs.min()),
-    "resized_image": img_base64,
-    "confirm_order": "offline_weights_loaded",
-  }
+  return {"predictions": results}

apps/api/main.py

@@ -1,12 +1,10 @@
-from fastapi import FastAPI, Request, Query
+from fastapi import FastAPI, Request
 from fastapi.responses import JSONResponse
 import logging
 import sys
 import traceback
 from aesthetic import score_aesthetic
 from classify import classify_image
-from PIL import Image
-import io
 
 # Setup logging once
 logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
@@ -30,22 +28,17 @@ def error_response(e: Exception):
     )
 
 @main_py_app.post("/classify")
-async def classify_endpoint(req: Request, debug: bool = Query(False)):
+async def classify_endpoint(req: Request):
     try:
-        results, debug_data = await classify_image(req, debug)
-        return {"predictions": results} if not debug else {
-            "predictions": results,
-            **debug_data
-        }
+        results = await classify_image(req)
+        return results
     except Exception as e:
         return error_response(e)
 
 @main_py_app.post("/scores")
 async def score_endpoint(req: Request):
     try:
-        img_bytes = await req.body()
-        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
-        score = score_aesthetic(img)
+        score = await score_aesthetic(req)
         return {"aesthetic_score": round(score, 3)}
     except Exception as e:
         return error_response(e)

apps/api/tests/test_aesthetic.py

@@ -1,9 +1,8 @@
-import pytest
 from PIL import Image
 from aesthetic import score_aesthetic
 
-def test_score_aesthetic_on_sample_image():
+async def test_score_aesthetic_on_sample_image():
     img = Image.new("RGB", (224, 224), color="blue")  # deterministic image
-    score = score_aesthetic(img)
+    score = await score_aesthetic(img)
     assert isinstance(score, float)
     assert -1.0 <= score <= 1.0

apps/api/tests/test_routes.py

@@ -37,6 +37,6 @@ def test_classify_route():
     print("DEBUG BODY:", body)
 
     assert "predictions" in body
-    predictions = body["predictions"]["predictions"]
+    predictions = body["predictions"]
     assert isinstance(predictions, list), f"Expected list, got: {type(predictions)}"
     assert all("label" in pred and "score" in pred for pred in predictions)