SushantGautam/Medico2025_subtask2_demo_submission

MEDICO 2025 — Task 2 Sample Submission (10 examples)

This archive demonstrates the submission format for Subtask 2 on 10 toy samples.
Replace placeholders with your actual outputs.

Files

• sample_submission_task2.jsonl — 10 rows, one per validation example.
• visuals/val_####_heatmap.png — example heatmaps referenced in visual_explanation.

JSONL Schema (per line)

{
  "val_id": 0,
  "img_id": "UNIQUE_IMAGE_IDENTIFIER",
  "question": "Original question posed to the model.",
  "answer": "Prediction from your model from Subtask 1 (exact string).",
  "textual_explanation": "Clinician-oriented reasoning referencing visual cues.",
  "visual_explanation": [{"type":"heatmap|segmentation_mask|bounding_box|etc.","data":"path/to.png|[[x1,y1,x2,y2]]","description":"(Optional) what it highlights"}],
  "confidence_score": 0.92
}

Field Requirements

• img_id / question / answer → Must match Subtask 1 data and predictions exactly.
• textual_explanation (Mandatory) → Use clinical language referencing location, morphology, color, size, vascular pattern, etc.
• visual_explanation (Optional but encouraged) → Heatmaps / masks / bounding boxes linked to the textual explanation.
• confidence_score (Optional but encouraged) → Float in [0, 1], estimated from your model (e.g., calibrated softmax, MC-dropout, temperature scaling).

How to use

1. Replace answer with your exact Subtask 1 predictions for these val_id/img_id pairs.
2. Update textual_explanation with a faithful, medically sound narrative that refers to the same ROI you ground visually.
3. If you produce masks or boxes, write them to visuals/ and point data to the relative path(s).
4. Keep confidence_score in [0, 1].

Notes

• This package does not contain actual Kvasir images or your model outputs—only a format template.
• Ensure val_id, img_id, and question match those from your val_set_task2.