|
Download README.md from 1779-lab/MMAD_remake_Benchmark: direct link, hf CLI and curl.
- Browser
- Download file 3.9 kB
-
https://huggingface.co/datasets/1779-lab/MMAD_remake_Benchmark/resolve/main/README.md
- Command line
-
hf download hf://datasets/1779-lab/MMAD_remake_Benchmark/README.md
-
curl -L -o README.md https://huggingface.co/datasets/1779-lab/MMAD_remake_Benchmark/resolve/main/README.md
3.9 kB
π MMAD Remake β Unified Industrial Anomaly Detection Benchmark
Welcome to the MMAD Remake Benchmark Dataset, a comprehensive, unified Multimodal Visual Question Answering (VQA) and detailed image captioning benchmark built for evaluating Frontier Vision-Language Models (VLMs) across industrial quality control and visual anomaly detection.
π Dataset Overview
- Combined Datasets (4 Sources):
- GoodsAD: Supermarket goods anomaly detection (4,875 images)
- MVTec AD: Industrial surface & object anomaly detection (5,354 images)
- VisA: High-resolution printed circuit boards & multi-instance objects (10,821 images)
- MVTec LOCO AD: Structural and logical constraint anomalies (3,122 images)
- Total Images: 24,172 high-resolution images
- Total VQA Question Set: ~120,000+ multiple-choice questions across 7 core industrial subtasks.
- Caption Model: Qwen2-VL-2B-Instruct
- VQA Generator Model: Qwen2.5-7B-Instruct (4-bit Quantized)
π Directory Structure
MMAD_remake_Benchmark/
βββ images/ # Contains image files organized by source dataset
β βββ goodsad/ # 6 supermarket product categories
β βββ mvtecad/ # 15 industrial object & texture categories
β βββ visa/ # 12 PCB & multi-instance categories
β βββ mvtecloco/ # 5 logical constraint categories
βββ mmad_remake_captions.json # Unified image captions across all 24,172 images
βββ mmad_remake_questions.json # Unified Multiple-Choice VQA Benchmark questions
βββ mmad_remake_metadata.csv # Overview metadata table for easy filtering
βββ README.md # Benchmark documentation
π Core VQA Subtasks Evaluated
- Anomaly Detection: Binary verification (
Yes/No) if a defect exists. - Anomaly Classification: Identifying the exact defect category/type.
- Anomaly Localization: Identifying the relative spatial location of the defect.
- Attribute QA: Inquiring about physical attributes (colors, OCR text, branding, material).
- Structural Reasoning: Evaluating component alignment, symmetry, and structural integrity.
- Logical Reasoning: Checking for missing parts, extra items, or placement constraint violations.
- Complex Decision Making: Final Quality Control pass/reject determination.
π οΈ Usage Example (Python PyTorch / HuggingFace)
import os
import json
import pandas as pd
from PIL import Image
BENCHMARK_ROOT = "./MMAD_remake_Benchmark"
QUESTIONS_FILE = os.path.join(BENCHMARK_ROOT, "mmad_remake_questions.json")
with open(QUESTIONS_FILE, "r", encoding="utf-8") as f:
benchmark_data = json.load(f)
print(f"Total benchmark samples: {len(benchmark_data)}")
# Load a sample item
sample = benchmark_data[0]
image_full_path = os.path.join(BENCHMARK_ROOT, sample["image_path"])
# Open image
img = Image.open(image_full_path)
print(f"Sample Global ID: {sample['global_id']} (Source: {sample['dataset_source']})")
print(f"Loaded image: {image_full_path} (Size: {img.size})")
# Iterate questions
for q in sample["questions"]:
print(f"\n[{q['subtask']}] {q['question']}")
for opt in q["options"]:
print(f" {opt}")
print(f"Correct Answer: {q['answer']}")
π License & Citation
The original images belong to their respective dataset creators (PKU-GoodsAD, MVTec Software GmbH, Amazon Research). Please cite the original papers when conducting non-commercial research using this benchmark:
@inproceedings{mmad_remake_2026,
title={MMAD Remake: A Unified Multimodal Benchmark for Industrial Visual Anomaly Detection},
author={Celesnity Research},
year={2026}
}