Update dataset card: add Warehouse_026/027 real-world test scenes, evaluation-script link, and 2026 quantification

README.md

@@ -109,6 +109,7 @@ Contains detailed calibration metadata per sensor:
 - **2024 Edition**: Evaluation based on 3D-location-based HOTA scores at the [2024 AI City Challenge Server](https://eval.aicitychallenge.org/aicity2024). The submission is currently disabled, as the ground truths of test set are provided with this release.
 - **2025 Edition**: Evaluation based on 3D-bounding-box-based HOTA scores at the [2025 AI City Challenge Server](https://eval.aicitychallenge.org/aicity2025).
 - **2026 Edition**: Evaluation based on 3D-bounding-box-based HOTA scores at the 2026 AI City Challenge Server (link to be announced).
+- **Evaluation script**: An offline HOTA evaluator that reproduces the official AICity MTMC (3D-bounding-box HOTA) metric is provided in the [`spatialai-data-utils` evaluation tools](https://github.com/NVIDIA-AI-Blueprints/video-search-and-summarization/tree/develop/libs/analytics/spatialai-data-utils/tools/evaluation) — see `evaluate_aicity_mtmc.py`.
 
 #### Scene ID Tables
 
@@ -177,8 +178,11 @@ Contains detailed calibration metadata per sensor:
 | Test  | 23       | Warehouse_023  |
 | Test  | 24       | Warehouse_024  |
 | Test  | 25       | Warehouse_025  |
+| Test  | 26       | Warehouse_026  |
+| Test  | 27       | Warehouse_027  |
 
-*Scenes 001 / 003 / 005 / 007 / 021 / 024 are **Cosmos Transfer 2.5 (CT2.5)** augmented re-renderings of 000 / 002 / 004 / 006 / 020 / 023 respectively (same scene & cameras, augmented RGB).*
+*Scenes 001 / 003 / 005 / 007 / 021 / 024 are **Cosmos Transfer 2.5 (CT2.5)** augmented re-renderings of 000 / 002 / 004 / 006 / 020 / 023 respectively (same scene & cameras, augmented RGB).*<br>
+*`Warehouse_026` and `Warehouse_027` are **real-world** test captures (4 and 7 cameras); the first 60 s (1800 frames) per camera is provided — videos + calibration only (ground truth, depth maps, and top-down maps withheld).*
 
 </td>
 </tr>
@@ -189,7 +193,9 @@ Contains detailed calibration metadata per sensor:
 |-------------------------|-------------------------------------------------------|--------|-------|---------|---------------------------------------------------------------|--------------|--------------|------------|------------|
 | **MTMC_Tracking_2024** | 2D bounding boxes, multi-camera tracking IDs | 90 | 212 | 953 | Person: 2,481 | 52M | 135M | No | 216.95 GB |
 | **MTMC_Tracking_2025** | 2D & 3D bounding boxes, multi-camera tracking IDs | 23 | 42 | 504 | Person: 292<br>Forklift: 13<br>NovaCarter: 28<br>Transporter: 23<br>FourierGR1T2: 6<br>AgilityDigit: 1<br>**Overall:** 363 | 8.9M | 73M | Yes | 3.31 TB (~70 GB without depth maps) |
-| **MTMC_Tracking_2026** | 2D & 3D bounding boxes, multi-camera tracking IDs | 26 | 28.5 | 342 | Person: 901<br>Forklift: 121<br>PalletTruck: 115 *(new)*<br>Transporter: 71<br>FourierGR1T2: 70<br>AgilityDigit: 69<br>NovaCarter: 32<br>**Overall:** 1,379 | 11.1M | 33.2M | Yes | 2.94 TB (~50 GB without depth maps) |
+| **MTMC_Tracking_2026** | 2D & 3D bounding boxes, multi-camera tracking IDs | 28 | 28.7 | 353 | Person: 901<br>Forklift: 121<br>PalletTruck: 115 *(new)*<br>Transporter: 71<br>FourierGR1T2: 70<br>AgilityDigit: 69<br>NovaCarter: 32<br>**Overall:** 1,379 | 11.1M | 33.2M | Yes | 2.94 TB (~50 GB without depth maps) |
+
+*`MTMC_Tracking_2026` scene / camera / camera-hour totals include the 2 real-world test scenes (`Warehouse_026`, `Warehouse_027`; first 60 s per camera). Their ground truth and depth maps are withheld, so the box counts and the "Depth Maps" column reflect the synthetic scenes only.*
 
 ## References
 Please cite the following papers when using this dataset:
@@ -227,6 +233,7 @@ NVIDIA believes Trustworthy AI is a shared responsibility and we have establishe
 Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
 
 ## Changelog
+- **2026-06-04**: Released 2 real-world test scenes for `MTMC_Tracking_2026` — `Warehouse_026` (4 cameras) and `Warehouse_027` (7 cameras), provided as 1080p / 30 FPS video covering the first 60 seconds (1800 frames) of each scene — the labeled evaluation window — with 2025-format `calibration.json`. Ground truth, depth maps, and top-down maps are withheld for these test scenes. An offline HOTA evaluator that reproduces the official AICity MTMC metric is available in the [`spatialai-data-utils` evaluation tools](https://github.com/NVIDIA-AI-Blueprints/video-search-and-summarization/tree/develop/libs/analytics/spatialai-data-utils/tools/evaluation) (`evaluate_aicity_mtmc.py`).
 - **2026-05-15**: Released 3 synthetic test sets for `MTMC_Tracking_2026`. The remaining real-world test sets are scheduled for release at the end of May. The evaluation system for the 2026 AI City Challenge Track 1 has also been postponed to the end of May. In addition, the 2025 AI City Challenge Track 1 evaluation system has been temporarily disabled and will be re-opened after the conclusion of the 2026 AI City Challenge.
 - **2026-04-29**: Initial release of `MTMC_Tracking_2026` — 26 scenes, 342 cameras, 28.5 camera-hours of 1080p video, 11.1M 3D boxes / 33.2M 2D boxes, with depth maps; calibration and GT aligned to the 2025 schema. 6 scenes (`Warehouse_001`, `_003`, `_005`, `_007`, `_021`, `_024`) are **Cosmos Transfer 2.5 (CT2.5) augmented** re-renderings of their source scenes (same scene/cameras/calibration/depth/GT, augmented RGB).
 - **2025-07-14**: Provided evaluation tools for both 2024 and 2025 datasets.