Update dataset card README

README.md

@@ -2,7 +2,8 @@
 license: mit
 pretty_name: FARM Aerial Radio Map Dataset
 task_categories:
-  - aerial radio map application
+  - image-to-image
+  - feature-extraction
 tags:
   - radio-map
   - aerial-networking
@@ -11,26 +12,20 @@ tags:
   - foundation-model
   - simulation
 size_categories:
-  - 26.4M
+  - 10M<n<100M
 ---
 
-# FARM Aerial Radio Map Dataset
-
-## Overview
-
-This repository releases the constructed aerial radio map datasets based on ARM-Omni for FARM training and evaluation. The datasets correspond to Table 2 of the FARM paper and include `D1-D10`, `P1`, `F1`, and `A1`.
-
-ARM-Omni is introduced in the paper as a large-scale aerial radio environment dataset. This Hugging Face release provides the ARM datasets used in the FARM experiments as downloadable scene archives. It is not a full ARM-Omni release.
-
 Paper:
 
 - https://arxiv.org/abs/2604.17362
 
-Code:
 
-- The FARM codebase will be released separately.
+# FARM Aerial Radio Map (ARM) Dataset
+
+## Overview
+
+This repository releases the constructed ARM datasets based on ARM-Omni for FARM training, in-domain evaluation (`D1-D10`), and zero-shot evaluation (`P1`, `F1`, and `A1`). The dataset coverage is summarized below:
 
-## Dataset Coverage
 
 | Dataset | Frequencies (GHz) | Max Rx Height (m) | Beamwidths | Map Grid Size | Volume |
 | --- | --- | ---: | --- | --- | ---: |
@@ -48,7 +43,7 @@ Code:
 | F1 | 2.6, 7.1 | 120 | 30°, 120°, Iso | 512 × 512 | 15000 |
 | A1 | 2.1, 3.3, 3.5, 4.9, 5.9 | 120 | 60° | 512 × 512 | 15000 |
 
-`D1-D10` are used for primary training and evaluation. `P1`, `F1`, and `A1` are used for zero-shot evaluation.
+A fuller ARM-Omni release is planned for a future update.
 
 ## Repository Layout
 
@@ -74,13 +69,13 @@ Scene IDs are not globally continuous across dataset groups. Users should rely o
 
 ## Data Format
 
-Each `.npy` file stores an aerial radio map tensor with shape:
+Each released raw `.npy` file stores an ARM volume with shape:
 
 ```text
-1 × H × W × W
+1 × 30 × 512 × 512
 ```
 
-The first dimension is the radio-map channel. `H` is the height axis. `W` follows the map grid size listed in Dataset Coverage. Most dataset groups use `512 × 512` spatial grids, while `P1` uses `256 × 256`.
+The dimensions correspond to one radio-map channel, 30 height layers, and a `512 × 512` spatial grid. During data loading, different input sizes can be obtained by selecting a subset of height layers and applying a spatial crop centered on the transmitter pixel. For example, this procedure is used to obtain the `256 × 256` crop for `P1`.
 
 ## Signal Encoding and Scene Metadata
 
@@ -92,19 +87,10 @@ The first dimension is the radio-map channel. `H` is the height axis. `W` follow
 <dataset>/<scene_id>/tx_positions_512.csv
 ```
 
-- For `P1`, the released map grid is `256 × 256`. When using transmitter locations recorded in the original `512 × 512` scene coordinate system, the `256 × 256` region is centered on the transmitter pixel. For convenience, the crop window can be computed as:
-
-```python
-def compute_centered_crop_window(center_row, center_col, height, width, crop_size=256):
-    crop_h = min(int(crop_size), int(height))
-    crop_w = min(int(crop_size), int(width))
-    start_y = max(0, min(int(center_row) - crop_h // 2, int(height) - crop_h))
-    start_x = max(0, min(int(center_col) - crop_w // 2, int(width) - crop_w))
-    return start_y, start_y + crop_h, start_x, start_x + crop_w
-```
+````md
 ## Citation
 
-If you use this dataset, please cite the FARM paper:
+If you use this dataset, please cite our paper:
 
 ```text
 S. Gao, J. Liang, Y. Yuan, W. Lu, G. Shen, and L. Yang, "FARM: Foundational Aerial Radio Map for Intelligent Low-Altitude Networking," arXiv preprint arXiv:2604.17362, 2026.