[HIRING!] 2026 Graduate Students & Undergraduate Interns (석박사 대학원생 및 학부인턴 모집)
Junha Shin JIwoong Jang and Hyunsuk Ko
This dataset consists of RGB, thermal, near-infrared (NIR), and depth data captured in indoor environments, together with spatially aligned thermal and NIR images, pseudo-depth maps, and valid-depth masks.
Indoor depth estimation remains challenging under conditions such as low illumination, weak texture, reflective surfaces, and modality-specific degradation. RGB images provide rich appearance and semantic information, while thermal and near-infrared images provide complementary information when visible-light imagery becomes unreliable.
The MIDD dataset provides paired multi-modal data captured in various indoor environments. Each sample contains RGB images, raw NIR images, raw 16-bit thermal data, sensor-acquired depth maps, pseudo-depth maps, and valid-depth masks. Spatially aligned NIR and thermal images are also provided in the RGB image coordinate system, enabling pixel-level multi-modal fusion and consistent depth estimation.
By providing both raw and spatially aligned multi-modal data together with depth supervision, MIDD serves as a benchmark dataset for multi-modal indoor depth estimation, cross-modal feature fusion, alignment-aware learning, and robustness analysis under diverse indoor conditions.
The MIDD dataset provides indoor multi-modal data composed of RGB, thermal, NIR, and depth information.
For each capture sequence, the dataset includes RGB images, raw NIR images, raw 16-bit thermal data, sensor-acquired depth maps, pseudo-depth maps, and valid-depth masks.
To support pixel-level multi-modal learning, spatially aligned NIR and thermal images are additionally provided in the RGB coordinate system as warped_nir and warped_thr, respectively.
The dataset is organized as follows:
MIDD/
└── raw_final/
└── sequence_name/
├── RGB/
├── NIR/
├── Thermal16_raw_npz/
├── warped_nir/
├── warped_thr/
├── depth/
├── pseudo_depth/
└── valid_mask/
To access the MIDD dataset, please complete the dataset request form using the link below.
After the form is submitted, the dataset download link will be automatically sent to the email address provided in the form.
Copyright © 2026 Intelligent Visual Media Laboratory (IVML).
The MIDD dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License.
The dataset may be used, shared, and adapted for research and other non-commercial purposes, provided that appropriate attribution is given to the Intelligent Visual Media Laboratory and any derivative work is distributed under the same license.
By using this dataset, you agree to cite the associated MIDD publication in any paper, report, presentation, or other work that uses this dataset.