{ "fetched_at": "2026-10-19T07:11:40Z", "source_url": "https://www.kaggle.com/competitions/aiml-general-championship/data", "pages": [ { "name": "data-description", "content": "#**Input Data**\n\n\nThe input data are dermoscopic lesion images in JPEG format.\n\nAll lesion images are named using the scheme ISIC_ .jpg , where is a 7-digit unique identifier. EXIF tags in the images have been removed; any remaining EXIF tags should not be relied upon to provide accurate metadata.\n\nThe lesion images come from the HAM10000 Dataset, and were acquired with a variety of dermatoscope types, from all anatomic sites (excluding mucosa and nails), from a historical sample of patients presented for skin cancer screening, from several different institutions. Images were collected with approval of the Ethics Review Committee of University of Queensland (Protocol-No. 2017001223) and Medical University of Vienna (Protocol-No. 1804/2017).\n\nThe distribution of disease states represent a modified \"real world\" setting whereby there are more benign lesions than malignant lesions, but an over-representation of malignancies.\n\n\n#**Ground Truth Provenance**\n\n\nAs detailed in the HAM10000 Dataset description, diagnosis ground truth were established by one of the following methods:\n\n* Histopathology\n* Reflectance confocal microscopy\n* Lesion did not change during digital dermatoscopic follow up over two years with at least three images\n* Consensus of at least three expert dermatologists from a single image\n* In all cases of malignancy, disease diagnoses were histopathologically confirmed.\n\n\n\n#**Supporting papers**\n\n(1) Robust Asymmetric Loss for Multi-Label Long-Tailed Learning ([Paper](https://arxiv.org/abs/2308.05542) / [Code](https://github.com/kalelpark/RAL))\n(2) Focal Loss for Dense Object Detection ([Paper](https://arxiv.org/abs/1708.02002v2) / [Code]https://github.com/unsky/focal-loss))\n(3) Asymmetric Loss For Multi-Label Classification ([Paper](https://arxiv.org/abs/2009.14119) / [Code](https://github.com/Alibaba-MIIL/ASL))\n(4) Simple and Robust Loss Design for Multi-Label Learning with Missing Labels ([Paper](https://arxiv.org/abs/2112.07368) / [Code](https://github.com/xinyu1205/robust-loss-mlml))\n\n\n\n## Files\n\n* **Training Images** - The training images are stored in \"KCDH2024_Training_Input_10K\" folder. All of the images are of .jpg type.\n* **KCDH2024_Training_GroundTruth.xlsx** - The spreadsheet containing the ground truth of the training images, it has 8 columns with the first being the image name in the \"KCDH2024_Training_Input_10K\" folder and the corresponding Boolean values, with a \"1\" means that the image belongs to the category of that particular column.\n* **sample_submission.csv** - a sample submission file in the correct format\n* **KCDH2024_Training_LesionGroupings.xlsx** - It is an additional information about the training images. You may or may not use it for building your model.\n\n## Columns\n\n* `id` - The ID or name of the Image as corresponded in the **KCDH2024_Training_GroundTruth.xlsx**\n* `prediction` - The predicted class of the image.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "We are thankful to Dr. Raghvendra as well as KCDH for their efforts in curating the dataset as well as the problem statement for this General Championship.\n\n#**Goal**\n\n\nSubmit automated predictions of disease classification within dermoscopic images.\nPossible disease categories are:\n\n\n1. [Melanoma (MEL)](https://dermoscopedia.org/Melanoma)\n2. [Melanocytic nevus (NV)](https://dermoscopedia.org/Benign_Melanocytic_lesions)\n3. [Basal cell carcinoma (BCC)](https://dermoscopedia.org/Basal_cell_carcinoma)\n4. [Actinic keratosis / Bowen’s disease (intraepithelial carcinoma) (AKIEC)](https://dermoscopedia.org/Actinic_keratosis_/_Bowen%27s_disease_/_keratoacanthoma_/_squamous_cell_carcinoma)\n5. [Benign keratosis (solar lentigo / seborrheic keratosis / lichen planus-like keratosis) (BKL)](https://dermoscopedia.org/Solar_lentigines_/_seborrheic_keratoses_/_lichen_planus-like_keratosis)\n6. [Dermatofibroma (DF)](https://dermoscopedia.org/Dermatofibromas)\n7. [Vascular lesion (VASC)](https://dermoscopedia.org/Vascular_lesions)\n\n\n#**Getting Started**\n\nPlease review the detailed guidelines in the Data, Rules, and Evaluation sections. Download the provided dataset, develop your model, and submit your predictions before the competition deadline. We encourage open discussion and collaboration among participants on the competition's discussion forum.\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "#**Goal Metric**\n\nThe metric: Balanced Accuracy Score\n\nThe balanced accuracy in binary and multiclass classification problems to deal with imbalanced datasets. It is defined as the average of recall obtained on each class.\n\n**This is the sole metric that will be considered to determine your ranking.**\n\n#**Other Metrics**\n\n\nParticipants will be ranked based only on the multiclass accuracy metric. However, for scientific completeness, predicted responses will also have the following metrics computed (comparing prediction vs. ground truth) for each image:\n\n##**Individual Category Metrics**\n\n* [sensitivity](https://en.wikipedia.org/wiki/Sensitivity_and_specificity#Sensitivity)\n* [specificity](https://en.wikipedia.org/wiki/Sensitivity_and_specificity#Specificity)\n* [accuracy](https://en.wikipedia.org/wiki/Accuracy_and_precision#In_binary_classification)\n* [area under the receiver operating characteristic curve (AUC)](https://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_the_curve)\n* [mean average precision](http://fastml.com/what-you-wanted-to-know-about-mean-average-precision/)\n* [F1 score](https://en.wikipedia.org/wiki/F1_score)\n* average AUC across all diagnoses\n* [positive predictive value (PPV)](https://en.wikipedia.org/wiki/Positive_and_negative_predictive_values)\n* [negative predictive value (NPV)](https://en.wikipedia.org/wiki/Positive_and_negative_predictive_values)", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "# Guidelines\n\nAll students from hostels would be eligible for participation, representing their parent hostels.\n**For Team Name strictly adhere to the following format [TeamLeader'sName_HostelNo.]**\n\n\n## Registration\n\n- Participants have to register for this contest in teams of 5.\n- Strict actions will be taken if anyone is found to take part in GCs from another hostel than the one assigned to them. \n- The participation in the GC is limited to a maximum of 5 teams.\n- Everyone interested in participating must fill the google form to register for the GC: \n- Registration Deadline: 06th March, 11:59 PM\n\nInstitute Technical Council reserves all rights to disqualify/penalize any individual based on its own judgment in case of any plagiarism or violation of integrity in the spirit of the contest. In any case of dispute, the decision of the council members will be final and binding.\n\n## Evaluation\n\nThe AIML GC is a high-prep GC with a total of 100 points. ( 70 Normal points + 15 Bonus points + 10 Additional Bonus + 5 Publicity )\n\n### Evaluation of Individual Score Each Participating team\n\n- The platform for the championship is Kaggle. There is only 1 final submission per team.\n- Predicted responses are scored using a normalized multi-class accuracy metric (balanced across categories). Tied positions will be broken using the area under the receiver operating characteristic curve (AUC) metric.\n- **Other Metrics:** Participants will be ranked based only on the multiclass accuracy metric. However, for scientific completeness, predicted responses will also have the following metrics computed (comparing prediction vs. ground truth) for each image:\n - sensitivity\n - specificity\n - accuracy\n - area under the receiver operating characteristic curve (AUC)\n - mean average precision\n\nThe participants can submit their code for the challenges anytime between 11:59 PM on 6th March to 11:59 PM on 16th February 2024. \nIf some participant is found to be of a username which was not already filled using the registration form, that participant’s score would be discarded.\n\n## Grading Scheme for Overall Hostel Points\n\n- Normal points- On the basis of multi-class accuracy metric, each team from each hostel will be graded some X (between 0 and 10 points) points. Let’s say a particular hostel participated with 3 teams, the normal points A= 7\\* ((X1+ X2+ X3)/3), basically 7 times average of value X for all teams from that hostel.\n- Hostels will be allotted B= 15 bonus points if value A for that hostel is greater than average A value among all hostels. For eg. If the average value of A among all hostels comes up to be 45, a hostel will be allotted 15 extra points if their A value is higher than 45.\n- Additional Bonus C= 10 points will be allotted to the hostels from which all the teams calculated the “Other metrics” mentioned in the previous section, this would also be based on the presentation which would be conducted by the professors and WnCC organizing team.\n\nExtra publicity marks C = 5 will be awarded to every hostel that fulfill both of the following points with proof mailed to us via the mailing thread latest by 06 March, 11:59 PM:\n- The poster for the GC along with a WhatsApp message attaching the GC rulebook and registration link should be publicized in at least one hostel WhatsApp group.\n- A mail should be sent to all the hostel residents via the GSec/Tech Co Webmail ID/ regarding the GC with the poster, rulebook and registration link for the GC attached. \n\nThe total score will be calculated as T=A+B+C. The hostel will be ranked according to their T-values.\n\n*Note: If individuals from the same hostel are found to be colluding in order to increase their score, the plagiarism detector on Kaggle will figure it out and in such a case, ITC reserves the right to disqualify the Hostel from AI/ML GC 2024. Hence, participants are advised to act responsibly.\n\n## Point System\n\nIf two or more hostels are tied for a position, both of them will be awarded the position points, and the next position removed. Example: A tie for the first position will result in both getting the points corresponding to the first position, and the next hostel will receive the points corresponding to the third position.\n\n## Division of GC points\n\nThe AI/ML GC is a high-prep GC, and X = 100 GC points will be awarded. The points will be distributed as mentioned in the general GC guidelines. \n- 5% of GC points for participation\n- 1st: 100% of 95% of GC points\n- 2nd: 80% of 95% of GC points \n- 3rd: 60% of 95% of GC points\n- 4th: 40% of 95% of GC points\n- 5th: 20% of 95% of GC points\n\n### Participation points\n\n- 1st: 100% of 5% of GC points (Maximum Participation)\n- 2nd: 80% of 5% of GC points\n- Others: 60% of 5% of GC points\n\n## Note on Kaggle Username\n\nIndividuals are required to submit code on Kaggle only through one Kaggle account and the account’s username must be entered in the registration form. Any submissions through other Kaggle accounts will not be considered.\n\n**For Team Name strictly adhere to the following format [TeamLeader'sName_HostelNo.]**\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "abstract", "content": "\n\nDermoscopic lesion imaging, like many diagnostic medical exams, produces a long-tailed distribution of clinical findings; while a small subset of diseases are routinely observed, the vast majority of diseases are relatively rare. This poses a challenge for standard AIML methods, which exhibit bias toward the most common classes at the expense of the important, but rare, “tail” classes. Many existing methods have been proposed to tackle this specific type of imbalance, though only recently with attention to long-tailed medical image recognition problems.\n\nDiagnosis on dermoscopic lesion images is a multi-class problem, as each image is associated with a single disease class. This dataset has 10, 015 skin images with 7 lesion classes. This dataset is characterized by an imbalance ratio (IR) of 58, where IR is defined as the ratio of the sample size of the largest majority class and that of the smallest minority class. Thus the larger the value of IR, the larger the imbalance extent.\n\n![lesion imaging](https://www.researchgate.net/profile/Aimilios-Lallas/publication/236336605/figure/fig4/AS:669293918298134@1536583513722/Dermoscopy-in-the-treatment-selection-and-follow-up-A-Clinically-this-superficial_Q640.jpg)\n\nSince most large-scale image classification benchmarks contain single-label images with a mostly balanced distribution of labels, many standard AIML methods fail to accommodate the class imbalance problem posed by the long-tailed nature of tasks like disease diagnosis on dermoscopic lesion images.\n\nTo develop a benchmark for long-tailed, multi-class medical image classification, we will use the HAM10000 Dataset that was acquired with a variety of dermatoscope types that are categorized into one of seven possible disease categories.\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Submission Format", "content": "For each ID (or name of the image) in the test set, you must predict the corresponding class. The submission file should be a CSV having the following format (header must be included):\n\nThe header should be the text written except the brackets.\n\n ID(name of the image),Class(number of image)\n image a,0\n image b,1\n image c,2\n image d,3\n image e,4\n image f,5\n image g,6\n\nHere, the numbers correspond to the classes in the same order as they are given in the ground truth file, which is:\n0 - MEL\n1 - NV\n2 - BCC\n3 - AKIEC\n4 - BKL\n5 - DF\n6 - VASC", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }