From 47c35da17e256f18d2fea3fce4c55e9b883f4d5f Mon Sep 17 00:00:00 2001 From: vshnosngh Date: Thu, 29 Oct 2026 13:27:57 +0530 Subject: [PATCH] feat(image-to-text): clean json export to image_file and prompt only, sort queue sequentially by filename, and auto-resize images >720px on upload --- .../image_to_prompt/processor.py | 44 ++-- cmd/image-to-text/main.go | 71 +++++- cmd/image-to-text/worker.py | 6 + web/image-to-text/app.js | 213 +++++++++++++++--- 4 files changed, 280 insertions(+), 54 deletions(-) diff --git a/cmd/image-to-text/image_to_prompt/processor.py b/cmd/image-to-text/image_to_prompt/processor.py index 7fa14e3..ff6a243 100644 --- a/cmd/image-to-text/image_to_prompt/processor.py +++ b/cmd/image-to-text/image_to_prompt/processor.py @@ -7,6 +7,7 @@ from __future__ import annotations import csv import json +import re import time from pathlib import Path from typing import Any, Callable, Generator, List, Optional, Union @@ -40,7 +41,7 @@ def find_image_files( if p.is_file() and p.suffix.lower() in valid_exts: image_files.append(p) - image_files.sort() + image_files.sort(key=lambda p: [int(c) if c.isdigit() else c.lower() for c in re.split(r'(\d+)', p.name)]) return image_files @@ -86,19 +87,29 @@ class ImageProcessor: "prompt": prompt_text, } - # Check image dimensions + # Check image dimensions & resize to 720 if greater + img_input = image_path if isinstance(image_path, (str, Path)): p = Path(image_path) if not p.is_file(): raise FileNotFoundError(f"Image not found: {p}") with Image.open(p) as img: image_info["width"], image_info["height"] = img.size + if max(img.size) > 720: + img_copy = img.convert("RGB") + img_copy.thumbnail((720, 720), Image.Resampling.LANCZOS) + img_input = img_copy + image_info["width"], image_info["height"] = img_copy.size elif isinstance(image_path, Image.Image): image_info["width"], image_info["height"] = image_path.size + if max(image_path.size) > 720: + image_path.thumbnail((720, 720), Image.Resampling.LANCZOS) + image_info["width"], image_info["height"] = image_path.size + img_input = image_path # Run inference generated_text = self.model.generate( - image_input=image_path, + image_input=img_input, prompt=prompt_text, max_new_tokens=max_new_tokens, temperature=temperature, @@ -248,26 +259,21 @@ class ImageProcessor: total_elapsed = time.time() - start_batch_time - # Save aggregated JSON + # Save aggregated JSON (strictly image_file and prompt only) if save_json: json_file = Path(save_json) json_file.parent.mkdir(parents=True, exist_ok=True) + clean_items = [ + { + "image_file": Path(item.get("file", "")).name if item.get("file") else "", + "prompt": item.get("result", "") + } + for item in results + if item.get("result") + ] with open(json_file, "w", encoding="utf-8") as f: - json.dump( - { - "total": total_images, - "processed": processed_count, - "skipped": skipped_count, - "failed": failed_count, - "mode": mode, - "elapsed_seconds": round(total_elapsed, 2), - "items": results, - }, - f, - indent=2, - ensure_ascii=False, - ) - print(f"[+] Aggregated JSON saved to: {json_file}") + json.dump(clean_items, f, indent=2, ensure_ascii=False) + print(f"[+] Aggregated clean JSON saved to: {json_file}") # Save aggregated CSV if save_csv: diff --git a/cmd/image-to-text/main.go b/cmd/image-to-text/main.go index 766a52f..2b14638 100644 --- a/cmd/image-to-text/main.go +++ b/cmd/image-to-text/main.go @@ -16,6 +16,7 @@ import ( "os/exec" "os/signal" "path/filepath" + "sort" "strconv" "strings" "sync" @@ -664,8 +665,28 @@ func handleExportZip(w http.ResponseWriter, r *http.Request) { } } - // 2. Write dataset summary JSON - summaryBytes, err := json.MarshalIndent(jobData, "", " ") + // 2. Write dataset summary JSON (strictly image_file and prompt only) + type CleanPromptJSON struct { + ImageFile string `json:"image_file"` + Prompt string `json:"prompt"` + } + var cleanSummary []CleanPromptJSON + for _, item := range jobData.Results { + filename, _ := item["filename"].(string) + if filename == "" { + file, _ := item["file"].(string) + filename = filepath.Base(file) + } + resultText, _ := item["result"].(string) + if resultText != "" { + cleanSummary = append(cleanSummary, CleanPromptJSON{ + ImageFile: filename, + Prompt: resultText, + }) + } + } + + summaryBytes, err := json.MarshalIndent(cleanSummary, "", " ") if err == nil { f, err := zipWriter.Create("dataset_summary.json") if err == nil { @@ -716,7 +737,22 @@ func handleExportZipDirect(w http.ResponseWriter, r *http.Request) { } } - summaryBytes, _ := json.MarshalIndent(req, "", " ") + // Write dataset summary JSON (strictly image_file and prompt only) + type CleanPromptJSON struct { + ImageFile string `json:"image_file"` + Prompt string `json:"prompt"` + } + var cleanSummary []CleanPromptJSON + for _, it := range req.Items { + if it.Text != "" { + cleanSummary = append(cleanSummary, CleanPromptJSON{ + ImageFile: it.Filename, + Prompt: it.Text, + }) + } + } + + summaryBytes, _ := json.MarshalIndent(cleanSummary, "", " ") if f, err := zipWriter.Create("dataset_summary.json"); err == nil { _, _ = f.Write(summaryBytes) } @@ -785,6 +821,11 @@ func handleScanFolder(w http.ResponseWriter, r *http.Request) { } } + // Sort files sequentially by filename + sort.Slice(files, func(i, j int) bool { + return files[i].Name < files[j].Name + }) + writeJSON(w, http.StatusOK, map[string]any{ "folder": folderPath, "total": len(files), @@ -800,9 +841,29 @@ func handleExportJSON(w http.ResponseWriter, r *http.Request) { return } + type CleanPromptJSON struct { + ImageFile string `json:"image_file"` + Prompt string `json:"prompt"` + } + var cleanResults []CleanPromptJSON + for _, item := range jobData.Results { + filename, _ := item["filename"].(string) + if filename == "" { + file, _ := item["file"].(string) + filename = filepath.Base(file) + } + resultText, _ := item["result"].(string) + if resultText != "" { + cleanResults = append(cleanResults, CleanPromptJSON{ + ImageFile: filename, + Prompt: resultText, + }) + } + } + w.Header().Set("Content-Type", "application/json") - w.Header().Set("Content-Disposition", fmt.Sprintf("attachment; filename=\"batch_results_%s.json\"", jobID[:8])) - _ = json.NewEncoder(w).Encode(jobData) + w.Header().Set("Content-Disposition", fmt.Sprintf("attachment; filename=\"dataset_prompts_%s.json\"", jobID[:8])) + _ = json.NewEncoder(w).Encode(cleanResults) } func handleExportCSV(w http.ResponseWriter, r *http.Request) { diff --git a/cmd/image-to-text/worker.py b/cmd/image-to-text/worker.py index aae72a0..4188eeb 100644 --- a/cmd/image-to-text/worker.py +++ b/cmd/image-to-text/worker.py @@ -178,6 +178,8 @@ def generate(req: GenerateRequest): raise HTTPException(status_code=404, detail=f"Image file not found: {req.image_path}") try: image = Image.open(img_path).convert("RGB") + if max(image.width, image.height) > 720: + image.thumbnail((720, 720), Image.Resampling.LANCZOS) except Exception as exc: raise HTTPException(status_code=400, detail=f"Failed to read image: {exc}") elif req.image_base64: @@ -187,6 +189,8 @@ def generate(req: GenerateRequest): raw_data = raw_data.split(",", 1)[1] img_bytes = base64.b64decode(raw_data) image = Image.open(io.BytesIO(img_bytes)).convert("RGB") + if max(image.width, image.height) > 720: + image.thumbnail((720, 720), Image.Resampling.LANCZOS) except Exception as exc: raise HTTPException(status_code=400, detail=f"Failed to decode base64 image: {exc}") else: @@ -289,6 +293,8 @@ def _execute_batch_job(job_id: str, req: BatchStartRequest): t0 = time.time() try: img = Image.open(img_path).convert("RGB") + if max(img.width, img.height) > 720: + img.thumbnail((720, 720), Image.Resampling.LANCZOS) text = model.generate( image_input=img, prompt=effective_prompt, diff --git a/web/image-to-text/app.js b/web/image-to-text/app.js index 4953cf5..2319571 100644 --- a/web/image-to-text/app.js +++ b/web/image-to-text/app.js @@ -237,38 +237,172 @@ // ── Unified Intake & Queue Management ── - function addFilesToQueue(files) { - const validExts = [".jpg", ".jpeg", ".png", ".webp", ".bmp", ".tiff", ".tif", ".avif"]; - let addedCount = 0; + // Helper: Naturally sort queue items sequentially by file name + function sortQueueByFilename() { + state.queue.sort((a, b) => + a.name.localeCompare(b.name, undefined, { numeric: true, sensitivity: "base" }) + ); + } - for (const file of files) { + // Helper: Resize image to max 720px right when uploaded if greater than 720px + function prepareAndResizeImage(file, maxDimension = 720) { + return new Promise((resolve) => { + const validExts = [".jpg", ".jpeg", ".png", ".webp", ".bmp", ".tiff", ".tif", ".avif"]; const ext = "." + file.name.split(".").pop().toLowerCase(); - if (!validExts.includes(ext) && !file.type.startsWith("image/")) continue; + if (!validExts.includes(ext) && !file.type.startsWith("image/")) { + resolve(null); + return; + } - const previewUrl = URL.createObjectURL(file); - state.queue.push({ - id: "q_" + Date.now() + "_" + Math.random().toString(36).slice(2, 7), - file: file, - serverPath: null, - name: file.name, - size: file.size, - previewUrl: previewUrl, - status: "pending", // "pending" | "processing" | "done" | "failed" - result: "", - elapsed: 0, - width: 0, - height: 0, - error: "" - }); - addedCount++; + const img = new Image(); + const objectUrl = URL.createObjectURL(file); + + img.onload = () => { + const origW = img.naturalWidth || img.width; + const origH = img.naturalHeight || img.height; + + // If within 720 on both sides, keep original file untouched + if (origW <= maxDimension && origH <= maxDimension) { + resolve({ + file: file, + previewUrl: objectUrl, + width: origW, + height: origH, + name: file.name, + size: file.size, + resized: false + }); + return; + } + + // If greater than 720, downscale preserving aspect ratio + let targetW = origW; + let targetH = origH; + if (origW >= origH) { + targetW = maxDimension; + targetH = Math.max(1, Math.round((origH * maxDimension) / origW)); + } else { + targetH = maxDimension; + targetW = Math.max(1, Math.round((origW * maxDimension) / origH)); + } + + const canvas = document.createElement("canvas"); + canvas.width = targetW; + canvas.height = targetH; + const ctx = canvas.getContext("2d", { alpha: false }); + ctx.imageSmoothingEnabled = true; + ctx.imageSmoothingQuality = "high"; + ctx.drawImage(img, 0, 0, targetW, targetH); + + let mimeType = file.type; + if (!mimeType || mimeType === "" || mimeType.includes("bmp") || mimeType.includes("tiff")) { + mimeType = "image/jpeg"; + } + + canvas.toBlob( + (blob) => { + URL.revokeObjectURL(objectUrl); + if (!blob) { + resolve({ + file: file, + previewUrl: URL.createObjectURL(file), + width: origW, + height: origH, + name: file.name, + size: file.size, + resized: false + }); + return; + } + + const resizedFile = new File([blob], file.name, { + type: blob.type || mimeType, + lastModified: Date.now() + }); + + const previewUrl = URL.createObjectURL(resizedFile); + resolve({ + file: resizedFile, + previewUrl: previewUrl, + width: targetW, + height: targetH, + name: file.name, + size: resizedFile.size, + resized: true, + origW: origW, + origH: origH + }); + }, + mimeType, + 0.92 + ); + }; + + img.onerror = () => { + URL.revokeObjectURL(objectUrl); + resolve(null); + }; + + img.src = objectUrl; + }); + } + + async function addFilesToQueue(files) { + if (!files || files.length === 0) return; + + const totalFiles = files.length; + if (totalFiles > 12) { + showToast(`Uploading & optimizing ${totalFiles} images (auto-resizing >720px)...`, "info"); + } + + let addedCount = 0; + let resizedCount = 0; + + // Process in concurrent batches of 6 for high performance & responsive UI + const batchSize = 6; + for (let i = 0; i < files.length; i += batchSize) { + const batch = files.slice(i, i + batchSize); + const preparedBatch = await Promise.all(batch.map((f) => prepareAndResizeImage(f, 720))); + + for (const prep of preparedBatch) { + if (!prep) continue; + if (prep.resized) resizedCount++; + + state.queue.push({ + id: "q_" + Date.now() + "_" + Math.random().toString(36).slice(2, 7), + file: prep.file, + serverPath: null, + name: prep.name, + size: prep.size, + previewUrl: prep.previewUrl, + status: "pending", // "pending" | "processing" | "done" | "failed" + result: "", + elapsed: 0, + width: prep.width, + height: prep.height, + resized: prep.resized, + error: "" + }); + addedCount++; + } } if (addedCount > 0) { - showToast(`Loaded ${addedCount} image(s)`, "info"); + // Sort queue sequentially by filename + sortQueueByFilename(); + + if (resizedCount > 0) { + showToast(`Loaded ${addedCount} images (${resizedCount} resized to max 720px)`, "info"); + } else { + showToast(`Loaded ${addedCount} image(s)`, "info"); + } updateQueueUI(); } } + window.addFilesToStudio = addFilesToQueue; + window.getStudioQueue = () => state.queue; + function addServerFilesToQueue(filesList) { let addedCount = 0; for (const f of filesList) { @@ -284,11 +418,14 @@ elapsed: 0, width: 0, height: 0, + resized: false, error: "" }); addedCount++; } if (addedCount > 0) { + // Sort queue sequentially by filename + sortQueueByFilename(); showToast(`Loaded ${addedCount} image(s) from server path`, "info"); updateQueueUI(); } @@ -797,6 +934,10 @@ const saveSidecar = saveSidecarCb.checked; const skipExisting = skipExistingCb.checked; + // Ensure queue is strictly sorted sequentially by filename before processing + sortQueueByFilename(); + updateQueueUI(); + const total = state.queue.length; // Process queue items sequentially so user sees immediate results per item @@ -946,12 +1087,18 @@ }; downloadTxtBtn.addEventListener("click", () => downloadSingleTxt()); - // 3. Single JSON - downloadJsonBtn.addEventListener("click", () => { - const item = state.queue[state.selectedIndex]; + // 3. Single JSON (Strictly image_file and prompt only) + window.downloadSingleJson = function (itemId) { + const item = state.queue.find((x) => x.id === itemId) || state.queue[state.selectedIndex]; if (!item || !item.result) return; - downloadBlob(JSON.stringify(item, null, 2), `${item.name.replace(/\.[^/.]+$/, "")}_output.json`, "application/json"); - }); + const stem = item.name.replace(/\.[^/.]+$/, ""); + const cleanJson = { + image_file: item.name, + prompt: item.result + }; + downloadBlob(JSON.stringify(cleanJson, null, 2), `${stem}.json`, "application/json"); + }; + downloadJsonBtn.addEventListener("click", () => downloadSingleJson()); // 4. Batch ZIP of All Sidecars (Direct Rock-solid Server Zip Generation) downloadAllZipBtn.addEventListener("click", async () => { @@ -1008,14 +1155,20 @@ downloadBlob(csv, `batch_results_${Date.now()}.csv`, "text/csv;charset=utf-8"); }); - // 6. Batch JSON Export + // 6. Batch JSON Export (Strictly image_file and prompt only) exportJsonBtn.addEventListener("click", () => { - const doneItems = state.queue.filter((x) => x.status === "done" || x.status === "failed"); + const doneItems = state.queue.filter((x) => x.status === "done" && x.result); if (doneItems.length === 0) { showToast("No processed results to export", "error"); return; } - downloadBlob(JSON.stringify(doneItems, null, 2), `batch_results_${Date.now()}.json`, "application/json"); + + const cleanJsonList = doneItems.map((it) => ({ + image_file: it.name, + prompt: it.result + })); + + downloadBlob(JSON.stringify(cleanJsonList, null, 2), `dataset_prompts_${Date.now()}.json`, "application/json"); }); // 7. Copy All Prompts