Update README.md

README.md

@@ -33,26 +33,31 @@ configs:
     path: data/location-*
   - split: placement
     path: data/placement-*
+
 ---
 
 
 
 <!-- New benchmark release announcement -->
+
 <div style="background-color: #ecfdf5; border-left: 4px solid #10b981; padding: 0.75em 1em; margin-top: 1em; color: #065f46; font-weight: bold; border-radius: 0.375em;">
   🎉 This repository contains the new version of <strong>RefSpatial-Bench</strong> — <strong>RefSpatial-Expand-Bench</strong>!<br>
   Based on the original benchmark, the new version <strong>extends indoor scenes</strong> (e.g., factories, stores) and adds <strong>previously uncovered outdoor scenarios</strong> (e.g., streets, parking lots), providing a more comprehensive evaluation of spatial referring tasks.
 </div>
 
+
 <div style="background-color: #fef3c7; border-left: 4px solid #f59e0b; padding: 0.75em 1em; margin-top: 1em; color: #78350f; font-weight: bold; border-radius: 0.375em;">
   🏆 The paper associated with this benchmark, <strong>RoboRefer</strong>, has been accepted to <strong>NeurIPS 2025</strong>!<br>
   Thank you all for your attention and support! 🙌
 </div>
 
 
+
 <h1 style="display: flex; align-items: center; justify-content: center; font-size: 1.75em; font-weight: 600;">
-  
+
+
   <img src="https://huggingface.co/datasets/BAAI/RefSpatial-Bench/resolve/main/assets/logo.png" style="height: 60px; flex-shrink: 0;">
-  
+
   <span style="line-height: 1.2; margin-left: 0px; text-align: center;">
     RefSpatial-Bench: A Benchmark for Multi-step Spatial Referring
   </span>
@@ -63,7 +68,7 @@ configs:
 <!-- # RefSpatial-Bench: A Benchmark for Multi-step Spatial Referring with Reasoning -->
 
  <!-- [![Generic badge](https://img.shields.io/badge/🤗%20Datasets-BAAI/RefSpatial--Bench-blue.svg)](https://huggingface.co/datasets/BAAI/RefSpatial-Bench)  -->
- 
+
 <p align="center">
   <a href="https://zhoues.github.io/RoboRefer"><img src="https://img.shields.io/badge/%F0%9F%8F%A0%20Project-Homepage-blue" alt="HomePage"></a>
   &nbsp;
@@ -77,6 +82,7 @@ configs:
 </p>
 
 
+
 Welcome to **RefSpatial-Bench**, a challenging benchmark based on real-world cluttered scenes to evaluate more complex multi-step spatial referring with reasoning.
 
 <img src="https://api.visitorbadge.io/api/combined?path=https%3A%2F%2Fzhoues.github.io&labelColor=%232ccce4&countColor=%230158f9" alt="visitor badge" style="display: none;" />
@@ -85,6 +91,7 @@ Welcome to **RefSpatial-Bench**, a challenging benchmark based on real-world clu
 
 
 <!-- ## 📝 Table of Contents
+
 * [🎯 Tasks](#🎯-tasks)
 * [🧠 Reasoning Steps](#🧠-reasoning-steps)
 * [📁 Dataset Structure](#📁-dataset-structure)
@@ -99,16 +106,12 @@ Welcome to **RefSpatial-Bench**, a challenging benchmark based on real-world clu
 * [📊 Dataset Statistics](#📊-dataset-statistics)
 * [🏆 Performance Highlights](#🏆-performance-highlights)
 * [📜 Citation](#📜-citation)
---- -->
+  --- -->
 
 ## 🎯 Task Split
-- Location Task: This task contains **100** samples, which requires model to predicts a 2D point indicating the **unique target object**.
-
-- Placement Task: This task contains **100** samples, which requires model to predicts a 2D point within the **desired free space**.
 
-- Unseen Set: This set comprises **77** samples from the Location/Placement task, specifically designed to **evaluate model generalization after SFT/RFT training on RefSpatial**, as it includes novel spatial relation combinations not present in RefSpatial.
-
-<div style="background-color: #ffe4e6; border-left: 4px solid #dc2626; padding: 0.75em 1em; margin-top: 1em; color: #b91c1c; font-weight: bold; border-radius: 0.375em;">   ⚠️ Warning: If your model is not trained with RefSpatial, Unseen set should not be used for evaluation. </div>
+- Location Task: This task contains **241** samples, which requires model to predicts a 2D point indicating the **unique target object**.
+- Placement Task: This task contains **200** samples, which requires model to predicts a 2D point within the **desired free space**.
 
 
 ## 🧠 Reasoning Steps
@@ -124,6 +127,7 @@ We provide two formats:
 <details>
 <summary><strong>Hugging Face Datasets Format</strong></summary>
 
+
 `data/` folder contains HF-compatible splits:
 
 * `location`
@@ -135,6 +139,7 @@ Each sample includes:
 | Field    | Description                                                  |
 | :------- | :----------------------------------------------------------- |
 | `id`     | Unique integer ID                                            |
+| scene    | indoor or outdoor                                            |
 | `object` | Natural language description of target (object or free area), which is extracted from the `prompt` |
 | `prompt` | Full Referring expressions                                   |
 | `suffix` | Instruction for answer formatting (**different models may use different suffixes or none**; we provide the format used by RoboRefer) |
@@ -147,6 +152,7 @@ Each sample includes:
 <details>
 <summary><strong>Raw Data Format</strong></summary>
 
+
 For full reproducibility and visualization, we also include the original files under:
 
 * `Location/`
@@ -173,9 +179,11 @@ Each entry in `question.json` has the following format:
   "rgb_path": "image/40.png",
   "mask_path": "mask/40.png",
   "category": "location",
-  "step": 2
+  "step": 2,
+  "scene": indoor
 }
 ```
+
 </details>
 
 
@@ -191,6 +199,7 @@ The following provides a quick guide on how to load and use the RefSpatial-Bench
 <details>
 <summary><strong>Method 1: Using Hugging Face Library</strong></summary>
 
+
 You can load the dataset easily using the `datasets` library:
 
 ```python
@@ -218,12 +227,14 @@ print(f"Prompt (from HF Dataset): {sample['prompt']}")
 print(f"Suffix (from HF Dataset): {sample['suffix']}")
 print(f"Reasoning Steps (from HF Dataset): {sample['step']}")
 ```
+
 </details>
 
 <details>
 <summary><strong>Method 2: Using Raw Data Files (JSON and Images)</strong></summary>
 
 
+
 If you are working with the raw data format (e.g., after cloning the repository or downloading the raw files), you can load the questions from the `question.json` file for each split and then load the images and masks using a library like Pillow (PIL).
 
 This example assumes you have the `location`, `placement`, and `unseen` folders (each containing `image/`, `mask/`, and `question.json`) in a known `base_data_path`.
@@ -272,17 +283,19 @@ if samples:
 else:
     print("No samples loaded.")
 ```
+
 </details>
 
 
 <details>
 <summary><strong>Evaluating RoboRefer / RoboPoint</strong></summary>
 
+
 To evaluate RoboRefer on RefSpatial-Bench:
 
 1. **Prepare Input Prompt:** 
 
-    Concatenate `sample["prompt"]` and `sample["suffix"]` to form the complete instruction.
+   Concatenate `sample["prompt"]` and `sample["suffix"]` to form the complete instruction.
 
    ```python
    # Example for constructing the full input for a sample
@@ -324,7 +337,7 @@ To evaluate RoboRefer on RefSpatial-Bench:
      # These scaled_roborefer_points are then used for evaluation against the mask.
      ```
 
-4. **Evaluation:** Compare `scaled_roborefer_points` against `sample["mask"]`. The main metric is **average success rate** — the percentage of predictions falling within the mask.
+3. **Evaluation:** Compare `scaled_roborefer_points` against `sample["mask"]`. The main metric is **average success rate** — the percentage of predictions falling within the mask.
 
 </details>
 
@@ -332,6 +345,7 @@ To evaluate RoboRefer on RefSpatial-Bench:
 <summary><strong>Evaluating Gemini Series</strong></summary>
 
 
+
 To evaluate Gemini Series on RefSpatial-Bench:
 
 1. **Prepare Input Prompt:** 
@@ -345,46 +359,47 @@ To evaluate Gemini Series on RefSpatial-Bench:
 
 2. **Model Prediction & JSON Parsing & Coordinate Scaling:** 
 
-     * **Model Prediction:** After providing the image (`sample["image"]`) and `full_input_instruction` to the Gemini model series, it outputs **normalized coordinates in an JSON format** like `"```json\n[\n  {\"point\": [y, x], \"label\": \"free space\"}, ...\n]\n```"`, where each `y` and `x` value is normalized to a range of 0-1000.
-
-     * **JSON Parsing:** Parse this JSON string to extract the coordinate attributes (e.g., `x1`, `y1`, `x2`, `y2`, etc.).
-
-     * **Coordinate Conversion:** To use these coordinates for evaluation against the mask, they must be:
-       
-       1.  Divided by 1000.0 to normalize them to the 0.0-1.0 range.
-       2.  Scaled to the original image dimensions (height for y, width for x).
-       ```python
-       # Example: model_output_gemini is "```json\n[\n  {\"point\": [438, 330], \"label\": \"free space\"}\n]\n```" from Gemini
-       # and sample["image"] is a PIL Image object loaded by the datasets library or loaded from the raw data
-       
-       def json2pts(text, width, height):
-          match = re.search(r"```(?:\w+)?\n(.*?)```", text, re.DOTALL)
-          if not match:
-              print("No valid code block found.")
-              return np.empty((0, 2), dtype=int)
+   * **Model Prediction:** After providing the image (`sample["image"]`) and `full_input_instruction` to the Gemini model series, it outputs **normalized coordinates in an JSON format** like `"```json\n[\n  {\"point\": [y, x], \"label\": \"free space\"}, ...\n]\n```"`, where each `y` and `x` value is normalized to a range of 0-1000.
+
+   * **JSON Parsing:** Parse this JSON string to extract the coordinate attributes (e.g., `x1`, `y1`, `x2`, `y2`, etc.).
+
+   * **Coordinate Conversion:** To use these coordinates for evaluation against the mask, they must be:
+
+     1.  Divided by 1000.0 to normalize them to the 0.0-1.0 range.
+     2.  Scaled to the original image dimensions (height for y, width for x).
+
+     ```python
+     # Example: model_output_gemini is "```json\n[\n  {\"point\": [438, 330], \"label\": \"free space\"}\n]\n```" from Gemini
+     # and sample["image"] is a PIL Image object loaded by the datasets library or loaded from the raw data
+     
+     def json2pts(text, width, height):
+        match = re.search(r"```(?:\w+)?\n(.*?)```", text, re.DOTALL)
+        if not match:
+            print("No valid code block found.")
+            return np.empty((0, 2), dtype=int)
       
-          json_cleaned = match.group(1).strip()
+        json_cleaned = match.group(1).strip()
       
-          try:
-              data = json.loads(json_cleaned)
-          except json.JSONDecodeError as e:
-              print(f"JSON decode error: {e}")
-              return np.empty((0, 2), dtype=int)
+        try:
+            data = json.loads(json_cleaned)
+        except json.JSONDecodeError as e:
+            print(f"JSON decode error: {e}")
+            return np.empty((0, 2), dtype=int)
       
-          points = []
-          for item in data:
-              if "point" in item and isinstance(item["point"], list) and len(item["point"]) == 2:
-                  y_norm, x_norm = item["point"]
-                  x = int(x_norm / 1000 * width)
-                  y = int(y_norm / 1000 * height)
-                  points.append((x, y))
+        points = []
+        for item in data:
+            if "point" in item and isinstance(item["point"], list) and len(item["point"]) == 2:
+                y_norm, x_norm = item["point"]
+                x = int(x_norm / 1000 * width)
+                y = int(y_norm / 1000 * height)
+                points.append((x, y))
       
-          return np.array(points)
-       
-       width, height = sample["image"].size 
-       scaled_gemini_points = json2pts(model_output_gemini, width, height)
-       # These scaled_gemini_points are then used for evaluation against the mask.
-       ```
+        return np.array(points)
+     
+     width, height = sample["image"].size 
+     scaled_gemini_points = json2pts(model_output_gemini, width, height)
+     # These scaled_gemini_points are then used for evaluation against the mask.
+     ```
 
 3. **Evaluation:** Compare `scaled_gemini_points` against `sample["mask"]`. The main metric is **average success rate** — the percentage of predictions falling within the mask.
 
@@ -393,6 +408,7 @@ To evaluate Gemini Series on RefSpatial-Bench:
 <details>
 <summary><strong>Evaluating the Molmo</strong></summary>
 
+
 To evaluate a Molmo model on this benchmark:
 
 1. **Prepare Input Prompt:** 
@@ -414,6 +430,7 @@ To evaluate a Molmo model on this benchmark:
 
      1.  Divide each coordinate by 100.0 to normalize it to the 0.0-1.0 range.
      2.  Scaled to the original image dimensions (height for y, width for x). 
+
      ```python
      # Example: model_output_molmo is '<points x1="61.5" y1="40.4" x2="76.8" y2="21.8"/>' from Molmo
      # and sample["image"] is a PIL Image object loaded by the datasets library or loaded from the raw data
@@ -431,45 +448,70 @@ To evaluate a Molmo model on this benchmark:
      ```
 
 3. **Evaluation:** Compare `scaled_molmo_points` against `sample["mask"]`. The main metric is **average success rate** — the percentage of predictions falling within the mask.
-</details>
+   </details>
 
 
 ## 📊 Dataset Statistics
 
 Detailed statistics on `step` distributions and instruction lengths are provided in the table below.
 
-| **RefSpatial-Bench** | **Step / Statistic** | **Samples** | **Avg. Prompt Length** |
-| :------------------- | :------------------- | :---------- | :--------------------- |
-| **Location**         | Step 1               | 30          | 11.13                  |
-|                      | Step 2               | 38          | 11.97                  |
-|                      | Step 3               | 32          | 15.28                  |
-|                      | **Avg. (All)**       | **100**     | 12.78                  |
-| **Placement**        | Step 2               | 43          | 15.47                  |
-|                      | Step 3               | 28          | 16.07                  |
-|                      | Step 4               | 22          | 22.68                  |
-|                      | Step 5               | 7           | 22.71                  |
-|                      | **Avg. (All)**       | **100**     | 17.68                  |
-| **Unseen**           | Step 2               | 29          | 17.41                  |
-|                      | Step 3               | 26          | 17.46                  |
-|                      | Step 4               | 17          | 24.71                  |
-|                      | Step 5               | 5           | 23.8                   |
-|                      | **Avg. (All)**       | **77**      | 19.45                  |
+| Task Type | Indoor  | Outdoor | Total   |
+| --------- | ------- | ------- | ------- |
+| Location  | 115     | 126     | 241     |
+| Placement | 120     | 80      | 200     |
+| **Total** | **235** | **206** | **441** |
+
+| Task Type | Step           | Samples | Avg. Prompt Length |
+| --------- | -------------- | ------- | ------------------ |
+| Location  | Step 1         | 54      | 10.61              |
+|           | Step 2         | 129     | 12.56              |
+|           | Step 3         | 58      | 16.10              |
+|           | **Avg. (All)** | **241** | **12.98**          |
+| Placement | Step 1         | 3       | 15.00              |
+|           | Step 2         | 86      | 15.14              |
+|           | Step 3         | 75      | 16.95              |
+|           | Step 4         | 29      | 22.24              |
+|           | Step 5         | 7       | 22.71              |
+|           | **Avg. (All)** | **200** | **17.11**          |
+
+
 
 ## 🏆 Performance Highlights
 
-As our research shows, **RefSpatial-Bench** presents a significant challenge to current models. In the table below, bold text indicates Top-1 accuracy, and underline text indicates Top-2 accuracy.
+Detailed accuracy results of RoboRefer-2B-SFT and RoboRefer-8B-SFT Models on RefSpatial-Bench-Expand
+
+#### **Location Task**
+
+| Category | 2B SFT | 8B SFT |
+| -------- | ------ | ------ |
+| Overall  | 50.21  | 61.00  |
+| Indoor   | 49.57  | 58.26  |
+| Outdoor  | 50.79  | 63.49  |
+| Step 1   | 61.11  | 72.22  |
+| Step 2   | 52.71  | 62.02  |
+| Step 3   | 34.48  | 48.28  |
+
+#### **Placement Task**
+
+| Category | 2B SFT | 8B SFT |
+| -------- | ------ | ------ |
+| Overall  | 48.50  | 60.00  |
+| Indoor   | 50.83  | 60.00  |
+| Outdoor  | 45.00  | 60.00  |
+| Step 1   | 33.33  | 33.33  |
+| Step 2   | 41.86  | 51.16  |
+| Step 3   | 54.67  | 70.67  |
+| Step 4   | 48.28  | 55.17  |
+| Step 5   | 71.43  | 85.71  |
+
 
-|   **Benchmark**    | **Gemini-2.5-Pro** | **SpaceLLaVA** | **RoboPoint** | **Molmo-7B** | **Molmo-72B** | **RoboRefer 2B-SFT** | **RoboRefer 8B-SFT** | **RoboRefer 2B-RFT** |
-| :----------------: | :----------------: | :------------: | :-----------: | :----------: | :-----------: | :------------: | :------------: | :------------: |
-| RefSpatial-Bench-L |    46.96           |      5.82      |     22.87     |    21.91     |     45.77     |  <u>47.00</u>  |   **52.00**    |   **52.00**    |
-| RefSpatial-Bench-P |       24.21        |      4.31      |     9.27      |    12.85     |     14.74     |     48.00      |   <u>53.00</u> |   **54.00**    |
-| RefSpatial-Bench-U |       27.14        |      4.02      |     8.40      |    12.23     |     21.24     |     33.77      |  <u>37.66</u>  |   **41.56**    |
 
 ## 📫 Contact
 
 If you have any questions about the benchmark, feel free to email Jingkun (anjingkun02@gmail.com) and Enshen (zhouenshen@buaa.edu.cn).
 <img src="https://api.visitorbadge.io/api/combined?path=https%3A%2F%2Fzhoues.github.io&labelColor=%232ccce4&countColor=%230158f9" alt="visitor badge" style="display: none;" />
 <img src="https://api.visitorbadge.io/api/combined?path=https%3A%2F%2Fanjingkun.github.io&labelColor=%232ccce4&countColor=%230158f9" alt="visitor badge" style="display: none;" />
+
 ## 📜 Citation
 
 Please consider citing our work if this benchmark is useful for your research.