{ "fetched_at": "2026-10-19T09:00:56Z", "source_url": "https://www.kaggle.com/competitions/bkai-igh-neopolyp/data", "pages": [ { "name": "data-description", "content": "## File description: \n- **test**: test images \n- **train**: training images \n- **train_gt**: ground-truths of training images \n- **sample_submission.csv**: sample submission \n\n## Ground-truth explanation \n- Red color: neoplastic polyps \n- Green color: non-neoplastic polyps \n- Black color: background ", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "

BKAI-IGH NeoPolyp-Small is a public dataset released by BKAI Research Center, Hanoi University of Science and Technology incorporation with Institute of Gastroenterology and Hepatology (IGH), Vietnam.

\n

This dataset contains 1200 images (1000 WLI and 200 FICE images). The training set consists of 1000 images, and the test set consists of 200 images. All polyps are classified into neoplastic or non-neoplastic classes denoted by red and green colors, respectively.

\n\n

A larger dataset called NeoPolyp-Large contains about 7500 images of four different color modes (WLI, BLI, LCI, FICE) with fire-grained annotations. Please check the information HERE for more details.

\n\nAll the images were collected in IGH. Annotations (including segmentation and classification) are added and verified by two experienced endoscopists from IGH.\n\n

Acknowledgments

\n

This dataset is collected thanks to the project VINIF.2020.DA17 \n funded by Vingroup Innovation Foundation. We thank IGH for collecting and annotating the data.

\n\n

References

\nIf you use this dataset in your work, please consider to cite the following papers:
\n1. Lan, P.N., An, N.S., Hang, D.V., Long, D.V., Trung, T.Q., Thuy, N.T., Sang, D.V.: NeoUnet: Towards accurate colon polyp segmentation and neoplasm detection. In: Proceedings of the 16th International Symposium on Visual Computing (2021)
\n2. Nguyen Thanh Duc, Nguyen Thi Oanh, Nguyen Thi Thuy, Tran Minh Triet, Dinh Viet Sang. ColonFormer: An Efficient Transformer Based Method for Colon Polyp Segmentation. IEEE Access, vol. 10, pp. 80575-80586, 2022
\n3. Nguyen Hoang Thuan, Nguyen Thi Oanh, Nguyen Thi Thuy, Perry Stuart, Dinh Viet Sang (2023). RaBiT: An Efficient Transformer using Bidirectional Feature Pyramid Network with Reverse Attention for Colon Polyp Segmentation. arXiv preprint arXiv:2307.06420.
\n4. Nguyen Sy An, Phan Ngoc Lan, Dao Viet Hang, Dao Van Long, Tran Quang Trung, Nguyen Thi Thuy, Dinh Viet Sang. BlazeNeo: Blazing fast polyp segmentation and neoplasm detection. IEEE Access, Vol. 10, 2022.\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "This competition is evaluated on the mean Dice coefficient. The Dice coefficient can be used to compare the pixel-wise agreement between a predicted segmentation and its corresponding ground truth. The formula is given by:\n$$ \\frac{2* |X \\cap Y|}{|X|+|Y|} $$\nwhere X is the predicted set of pixels and Y is the ground truth. The Dice coefficient is defined to be 1 when both X and Y are empty. The leaderboard score is the mean of the Dice coefficients for each ** pair in the test set. ClassId = 0 means neoplastic polyp, while ClassId = 1 means non-neoplastic polyp.\n\n**Expected results**\nIn order to reduce the submission file size, our metric uses run-length encoding on the pixel values. Instead of submitting an exhaustive list of indices for your segmentation, you will submit pairs of values that contain a start position and a run length. For example, '1 3' implies starting at pixel 1 and running a total of 3 pixels 1,2,3.\n\nThe competition format requires a space-delimited list of pairs. For example, '1 3 10 5' implies pixels 1,2,3,10,11,12,13,14 are to be included in the mask. The metric checks that the pairs are sorted, positive, and the decoded pixel values are not duplicated. The pixels are numbered from top to bottom, then left to right: 1 is pixel 1,1, 2 is pixel 2,1, etc.\n\n**File Format**\nYour submission file should be in csv format, with a header and columns names: Id, Expected.\nEach row in your submission represents a single predicted polyp segmentation for the given ImageId and predicted ClassId, and you should have the same number of rows as num_images * 2, i.e., 400 rows. The segment for each polyp class will be encoded into a single row, even if there are several non-contiguous polyp regions on an image.\n\n```\nId,Expected \n633a8d5b2b2b55157b7781e2c706c75c_0, 200 10 \n633a8d5b2b2b55157b7781e2c706c75c_1, 500 11 \n60a633a8d5b2b2b55157b7781e2c706c_0, 650 15 \n60a633a8d5b2b2b55157b7781e2c706c_1, 1 50 \netc...\n```\nYou can use the [sample code here](https://github.com/sangdv/rle_encode/blob/main/mask2csv.py) to convert segmentation masks into a csv file.\n\n**Important Note**\nIt should be emphasized that the predicted segmentation results must match the size of the original image. Therefore, if you resize the input image to fit your models, you must also resize the prediction output back to the original size of the input image before converting it to the required format.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }