The ORBIT (Object Recognition for Blind Image Training)-India Dataset
Bibliographic record
Abstract
The ORBIT (Object Recognition for Blind Image Training) -India Dataset is a collection of 105,243 images of 76 commonly used objects, collected by 12 individuals in India who are blind or have low vision. This dataset is an "Indian subset" of the original ORBIT dataset [1, 2], which was collected in the UK and Canada. In contrast to the ORBIT dataset, which was created in a Global North, Western, and English-speaking context, the ORBIT-India dataset features images taken in a low-resource, non-English-speaking, Global South context, a home to 90% of the world’s population of people with blindness. Since it is easier for blind or low-vision individuals to gather high-quality data by recording videos, this dataset, like the ORBIT dataset, contains images (each sized 224x224) derived from 587 videos. These videos were taken by our data collectors from various parts of India using the Find My Things [3] Android app. Each data collector was asked to record eight videos of at least 10 objects of their choice. Collected between July and November 2023, this dataset represents a set of objects commonly used by people who are blind or have low vision in India, including earphones, talking watches, toothbrushes, and typical Indian household items like a belan (rolling pin), and a steel glass. These videos were taken in various settings of the data collectors' homes and workspaces using the Find My Things Android app. The image dataset is stored in the ‘Dataset’ folder, organized by folders assigned to each data collector (P1, P2, ...P12) who collected them. Each collector's folder includes sub-folders named with the object labels as provided by our data collectors. Within each object folder, there are two subfolders: ‘clean’ for images taken on clean surfaces and ‘clutter’ for images taken in cluttered environments where the objects are typically found. The annotations are saved inside a ‘Annotations’ folder containing a JSON file per video (e.g., P1--coffee mug--clean--231220_084852_coffee mug_224.json) that contains keys corresponding to all frames/images in that video (e.g., "P1--coffee mug--clean--231220_084852_coffee mug_224--000001.jpeg": {"object_not_present_issue": false, "pii_present_issue": false}, "P1--coffee mug--clean--231220_084852_coffee mug_224--000002.jpeg": {"object_not_present_issue": false, "pii_present_issue": false}, ...). The ‘object_not_present_issue’ key is True if the object is not present in the image, and the ‘pii_present_issue’ key is True, if there is a personally identifiable information (PII) present in the image. Note, all PII present in the images has been blurred to protect the identity and privacy of our data collectors. This dataset version was created by cropping images originally sized at 1080 × 1920; therefore, an unscaled version of the dataset will follow soon. This project was funded by the Engineering and Physical Sciences Research Council (EPSRC) Industrial ICASE Award with Microsoft Research UK Ltd. as the Industrial Project Partner. We would like to acknowledge and express our gratitude to our data collectors for their efforts and time invested in carefully collecting videos to build this dataset for their community. The dataset is designed for developing few-shot learning algorithms, aiming to support researchers and developers in advancing object-recognition systems. We are excited to share this dataset and would love to hear from you if and how you use this dataset. Please feel free to reach out if you have any questions, comments or suggestions. REFERENCES: Daniela Massiceti, Lida Theodorou, Luisa Zintgraf, Matthew Tobias Harris, Simone Stumpf, Cecily Morrison, Edward Cutrell, and Katja Hofmann. 2021. ORBIT: A real-world few-shot dataset for teachable object recognition collected from people who are blind or low vision. DOI: https://doi.org/10.25383/city.14294597 microsoft/ORBIT-Dataset. https://github.com/microsoft/ORBIT-Dataset Linda Yilin Wen, Cecily Morrison, Martin Grayson, Rita Faia Marques, Daniela Massiceti, Camilla Longden, and Edward Cutrell. 2024. Find My Things: Personalized Accessibility through Teachable AI for People who are Blind or Low Vision. In Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems (CHI EA '24). Association for Computing Machinery, New York, NY, USA, Article 403, 1–6. https://doi.org/10.1145/3613905.3648641
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.005 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.006 | 0.003 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.014 | 0.026 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".