Update MMSearch-Plus dataset with encrypted data files

README.md

@@ -0,0 +1,161 @@
+---
+task_categories:
+- question-answering
+- visual-question-answering
+language:
+- en
+tags:
+- Multimodal Search
+- Multimodal Long Context
+size_categories:
+- n<1K
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: "*.arrow"
+dataset_info:
+  features:
+  - name: question
+    dtype: string
+  - name: answer
+    sequence: string
+  - name: num_images
+    dtype: int64
+  - name: arxiv_id
+    dtype: string
+  - name: video_url
+    dtype: string
+  - name: category
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: subtask
+    dtype: string
+  - name: img_1
+    dtype: image
+  - name: img_2
+    dtype: image
+  - name: img_3
+    dtype: image
+  - name: img_4
+    dtype: image
+  - name: img_5
+    dtype: image
+  splits:
+  - name: train
+    num_examples: 311
+---
+# MMSearch-Plus✨: Benchmarking Provenance-Aware Search for Multimodal Browsing Agents
+
+Official repository for the paper "[MMSearch-Plus: Benchmarking Provenance-Aware Search for Multimodal Browsing Agents](https://arxiv.org/abs/2508.21475)".
+
+🌟 For more details, please refer to the project page with dataset exploration and visualization tools: [https://mmsearch-plus.github.io/](https://mmsearch-plus.github.io).
+
+
+[[🌐 Webpage](https://mmsearch-plus.github.io/)] [[📖 Paper](https://arxiv.org/pdf/2508.21475)] [[🤗 Huggingface Dataset](https://huggingface.co/datasets/Cie1/MMSearch-Plus)] [[🏆 Leaderboard](https://mmsearch-plus.github.io/#leaderboard)]
+
+
+## 💥 News
+
+- **[2025.09.26]** 🔥 We update the [arXiv paper](https://arxiv.org/abs/2508.21475) and release all MMSearch-Plus data samples in [huggingface dataset](https://huggingface.co/datasets/Cie1/MMSearch-Plus).
+- **[2025.08.29]** 🚀 We release the [arXiv paper](https://arxiv.org/abs/2508.21475).
+
+## 📌 ToDo
+
+- Agentic rollout framework code
+- Evaluation script
+- Set-of-Mark annotations
+
+## Usage
+
+**⚠️ Important: This dataset is encrypted to prevent data contamination. However, decryption is handled transparently by the dataset loader.**
+
+### Simple Usage (Recommended)
+
+Load the dataset with automatic decryption using your canary string:
+
+```python
+import os
+from datasets import load_dataset
+
+# Set the canary string (hint: it's the name of this repo)
+os.environ['MMSEARCH_CANARY'] = 'your_canary_string'
+
+# Load dataset with transparent decryption
+dataset = load_dataset("Cie1/MMSearch-Plus", trust_remote_code=True)
+
+# Explore the dataset - everything is already decrypted!
+print(f"Dataset size: {len(dataset['train'])}")
+print(f"Features: {list(dataset['train'].features.keys())}")
+
+# Access a sample
+sample = dataset['train'][0]
+print(f"Question: {sample['question']}")
+print(f"Answer: {sample['answer']}")
+print(f"Category: {sample['category']}")
+print(f"Number of images: {sample['num_images']}")
+
+# Access images (PIL Image objects)
+sample['img_1'].show()  # Display the first image
+```
+
+## 👀 About MMSearch-Plus
+
+MMSearch-Plus is a challenging benchmark designed to test multimodal browsing agents' ability to perform genuine visual reasoning. Unlike existing benchmarks where many tasks can be solved with text-only approaches, MMSearch-Plus requires models to extract and use fine-grained visual cues through iterative image-text retrieval.
+
+### Key Features
+
+🔍 **Genuine Multimodal Reasoning**: 311 carefully curated tasks that cannot be solved without visual understanding
+
+🎯 **Fine-grained Visual Analysis**: Questions require extracting spatial cues and temporal traces from images to find out-of-image facts like events, dates, and venues
+
+🛠️ **Agent Framework**: Model-agnostic web agent with standard browsing tools (text search, image search, zoom-in)
+
+📍 **Set-of-Mark (SoM) Module**: Enables provenance-aware cropping and targeted searches with human-verified bounding box annotations
+
+### Dataset Structure
+
+Each sample contains:
+- Quuestion text and images
+- Ground truth answers and alternative valid responses
+- Metadata including arXiv id (if an event is a paper), video URL (if an event is a video), area and subfield
+
+### Performance Results
+
+Evaluation of closed- and open-source MLLMs shows:
+- Best accuracy is achieved by o3 with full rollout: **36.0%** (indicating significant room for improvement)
+- SoM integration provides consistent gains up to **+3.9 points**
+- Models struggle with multi-step visual reasoning and cross-modal information integration 
+
+<p align="center">
+    <img src="https://raw.githubusercontent.com/mmsearch-plus/mmsearch-plus.github.io/main/static/images/teaser.png" width="75%"> <br>
+    The overview of three paradigms for multimodal browsing tasks that demand fine-grained visual reasoning.
+</p>
+
+
+
+<p align="center">
+    <img src="https://raw.githubusercontent.com/mmsearch-plus/mmsearch-plus.github.io/main/static/images/real-teaser.jpg" width="60%"> <br>
+    The overview of an example trajectory for a task in <b>MMSearch-Plus</b>.
+</p>
+
+## 🏆 Leaderboard
+
+### Contributing to the Leaderboard
+
+🚨 The [Leaderboard](https://mmsearch-plus.github.io/#leaderboard) is continuously being updated, welcoming the contribution of your excellent LMMs!
+
+
+## 🔖 Citation
+
+If you find **MMSearch-Plus** useful for your research and applications, please kindly cite using this BibTeX:
+
+```latex
+@article{tao2025mmsearch,
+  title={MMSearch-Plus: A Simple Yet Challenging Benchmark for Multimodal Browsing Agents},
+  author={Tao, Xijia and Teng, Yihua and Su, Xinxing and Fu, Xinyu and Wu, Jihao and Tao, Chaofan and Liu, Ziru and Bai, Haoli and Liu, Rui and Kong, Lingpeng},
+  journal={arXiv preprint arXiv:2508.21475},
+  year={2025}
+}
+```

data-00000-of-00003.arrow

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d303f1ca1a9cd9470d401a538f55e1d4d70f9ca07aed8b2ab8d23f635e59831a
+size 419738728

data-00001-of-00003.arrow

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f070af1555340c0202102f292ef2f7920fccce0434b65c6fd384de407db9b9da
+size 466499832

data-00002-of-00003.arrow

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d59c12bdf310ad6399d4f94fc9cda5cabcd3a059e528accb8653a253e466adc
+size 345386360

dataset_info.json

@@ -0,0 +1,63 @@
+{
+  "citation": "",
+  "description": "",
+  "features": {
+    "question": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "answer": {
+      "feature": {
+        "dtype": "string",
+        "_type": "Value"
+      },
+      "_type": "Sequence"
+    },
+    "num_images": {
+      "dtype": "int64",
+      "_type": "Value"
+    },
+    "arxiv_id": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "video_url": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "category": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "difficulty": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "subtask": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "img_1": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "img_5": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "img_4": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "img_2": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "img_3": {
+      "dtype": "string",
+      "_type": "Value"
+    }
+  },
+  "homepage": "",
+  "license": ""
+}

mmsearch_plus.py

@@ -0,0 +1,166 @@
+"""MMSearch-Plus dataset with transparent decryption."""
+
+import base64
+import hashlib
+import io
+import os
+from typing import Dict, Any, List
+import datasets
+from PIL import Image
+
+_CITATION = """\
+@article{tao2025mmsearch,
+  title={MMSearch-Plus: A Simple Yet Challenging Benchmark for Multimodal Browsing Agents},
+  author={Tao, Xijia and Teng, Yihua and Su, Xinxing and Fu, Xinyu and Wu, Jihao and Tao, Chaofan and Liu, Ziru and Bai, Haoli and Liu, Rui and Kong, Lingpeng},
+  journal={arXiv preprint arXiv:2508.21475},
+  year={2025}
+}
+"""
+
+_DESCRIPTION = """\
+MMSearch-Plus is a challenging benchmark designed to test multimodal browsing agents' ability to perform genuine visual reasoning.
+Unlike existing benchmarks where many tasks can be solved with text-only approaches, MMSearch-Plus requires models to extract
+and use fine-grained visual cues through iterative image-text retrieval.
+"""
+
+_HOMEPAGE = "https://mmsearch-plus.github.io/"
+
+_LICENSE = "CC BY-NC 4.0"
+
+_URLS = {
+    "train": [
+        "data-00000-of-00003.arrow",
+        "data-00001-of-00003.arrow",
+        "data-00002-of-00003.arrow"
+    ]
+}
+
+def derive_key(password: str, length: int) -> bytes:
+    """Derive encryption key from password using SHA-256."""
+    hasher = hashlib.sha256()
+    hasher.update(password.encode())
+    key = hasher.digest()
+    return key * (length // len(key)) + key[: length % len(key)]
+
+def decrypt_image(ciphertext_b64: str, password: str) -> Image.Image:
+    """Decrypt base64-encoded encrypted image bytes back to PIL Image."""
+    if not ciphertext_b64:
+        return None
+
+    try:
+        encrypted = base64.b64decode(ciphertext_b64)
+        key = derive_key(password, len(encrypted))
+        decrypted = bytes([a ^ b for a, b in zip(encrypted, key)])
+
+        # Convert bytes back to PIL Image
+        img_buffer = io.BytesIO(decrypted)
+        image = Image.open(img_buffer)
+        return image
+    except Exception:
+        return None
+
+def decrypt_text(ciphertext_b64: str, password: str) -> str:
+    """Decrypt base64-encoded ciphertext using XOR cipher with derived key."""
+    if not ciphertext_b64:
+        return ciphertext_b64
+
+    try:
+        encrypted = base64.b64decode(ciphertext_b64)
+        key = derive_key(password, len(encrypted))
+        decrypted = bytes([a ^ b for a, b in zip(encrypted, key)])
+        return decrypted.decode('utf-8')
+    except Exception:
+        return ciphertext_b64
+
+class MmsearchPlus(datasets.GeneratorBasedBuilder):
+    """MMSearch-Plus dataset with transparent decryption."""
+
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        features = datasets.Features({
+            "question": datasets.Value("string"),
+            "answer": datasets.Sequence(datasets.Value("string")),
+            "num_images": datasets.Value("int64"),
+            "arxiv_id": datasets.Value("string"),
+            "video_url": datasets.Value("string"),
+            "category": datasets.Value("string"),
+            "difficulty": datasets.Value("string"),
+            "subtask": datasets.Value("string"),
+            "img_1": datasets.Image(),
+            "img_2": datasets.Image(),
+            "img_3": datasets.Image(),
+            "img_4": datasets.Image(),
+            "img_5": datasets.Image(),
+        })
+
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        # Get canary from environment variable or kwargs
+        canary = os.environ.get("MMSEARCH_CANARY")
+
+        # Check if passed in the builder's initialization
+        if hasattr(self, 'canary'):
+            canary = self.canary
+
+        if not canary:
+            raise ValueError(
+                "Canary string is required for decryption. Either set the MMSEARCH_CANARY "
+                "environment variable or pass it via the dataset loading kwargs. "
+                "Example: load_dataset('path/to/dataset', trust_remote_code=True) after setting "
+                "os.environ['MMSEARCH_CANARY'] = 'your_canary_string'"
+            )
+
+        # Download files
+        urls = _URLS["train"]
+        downloaded_files = dl_manager.download(urls)
+
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepaths": downloaded_files,
+                    "canary": canary,
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepaths, canary):
+        """Generate examples with transparent decryption."""
+        key = 0
+
+        for filepath in filepaths:
+            # Load the arrow file
+            arrow_dataset = datasets.Dataset.from_file(filepath)
+
+            for idx in range(len(arrow_dataset)):
+                example = arrow_dataset[idx]
+
+                # Decrypt text fields
+                if example.get("question"):
+                    example["question"] = decrypt_text(example["question"], canary)
+
+                if example.get("answer"):
+                    decrypted_answers = []
+                    for answer in example["answer"]:
+                        if answer:
+                            decrypted_answers.append(decrypt_text(answer, canary))
+                        else:
+                            decrypted_answers.append(answer)
+                    example["answer"] = decrypted_answers
+
+                # Decrypt image fields
+                image_fields = ["img_1", "img_2", "img_3", "img_4", "img_5"]
+                for field in image_fields:
+                    if example.get(field):
+                        example[field] = decrypt_image(example[field], canary)
+
+                yield key, example
+                key += 1

state.json

@@ -0,0 +1,19 @@
+{
+  "_data_files": [
+    {
+      "filename": "data-00000-of-00003.arrow"
+    },
+    {
+      "filename": "data-00001-of-00003.arrow"
+    },
+    {
+      "filename": "data-00002-of-00003.arrow"
+    }
+  ],
+  "_fingerprint": "f322dd0e15bea130",
+  "_format_columns": null,
+  "_format_kwargs": {},
+  "_format_type": null,
+  "_output_all_columns": false,
+  "_split": null
+}