REG 2025 Challenge
REG 2025 Challenge
1. Latest System Notices
[May 2025] Malware cleanup completed – servers restored, data re-uploaded.
Deadlines extended by one week – see the Important Dates section below.
Color-corrected TIFFs are available in
/REG_train_revised
(servers 104–107).Corrupted images were removed; full list in
corrupted_id.csv
.

Click the image to visit the challenge homepage
2. What’s New
UpdateImpact on Your WorkflowPre-trained weights ✅Any off-the-shelf vision or language backbone may be used.Extra data 🚫Outside training sets are banned to preserve fairness.Debug Phase 🔍Submit a sample run to verify output format before evaluation.Additional servers 📡Three extra mirrors added to ease download congestion.
3. Dataset at a Glance
20,500 WSIs from six centers in Korea, Japan, India, Turkey, and Germany.
Slide-level labels + original draft reports.
Average file size ≈ 1.2 GB; total > 24 TB.
Revised color profiles applied to 300+ slides (see
revised_list.csv
).
4. Evaluation Focus
Clinical accuracy over BLEU / ROUGE.
Real-world diagnostic value ranked by expert pathologists.
Bonus for concise, well-structured sections (macro / micro findings, impressions).
5. Recommended Workflow
Patch extraction + H&E normalization
ViT-L or SAM-ViT encoder fine-tuned on color-corrected slides
Seq-to-Seq decoder conditioned on slide embeddings + prior draft text
Rule-based medical NER layer to normalise terminology
Continuous evaluation with the provided GitHub script
6. Revised Schedule (KST 10 AM)
MilestoneOldNew (final)Training data release13 May 202520 May 2025Registration deadline27 Jun 202504 Jul 2025Debug Phase27 Jun → 03 Jul04 Jul → 10 JulTest 1 opens11 Jul 2025unchangedTest 1 deadline18 Jul → 25 Jul01 Aug 2025Test 2 opens02 Aug 2025unchangedTest 2 deadline09 Aug → 16 Aug23 Aug 2025Winners announced—09 Sep 2025
7. Quick Links
Registration & rules:
https://reg2025.grand-challenge.org/
Training dataset mirrors: Reg2025-TrainDataset page
Evaluation toolkit: GitHub → Submission & Evaluation
8. Final Checklist
Download revised TIFFs or run
tiffset
locally.Validate output JSON in Debug Phase.
Monitor further notices in the Updates channel.
1. Latest System Notices
[May 2025] Malware cleanup completed – servers restored, data re-uploaded.
Deadlines extended by one week – see the Important Dates section below.
Color-corrected TIFFs are available in
/REG_train_revised
(servers 104–107).Corrupted images were removed; full list in
corrupted_id.csv
.

Click the image to visit the challenge homepage
2. What’s New
UpdateImpact on Your WorkflowPre-trained weights ✅Any off-the-shelf vision or language backbone may be used.Extra data 🚫Outside training sets are banned to preserve fairness.Debug Phase 🔍Submit a sample run to verify output format before evaluation.Additional servers 📡Three extra mirrors added to ease download congestion.
3. Dataset at a Glance
20,500 WSIs from six centers in Korea, Japan, India, Turkey, and Germany.
Slide-level labels + original draft reports.
Average file size ≈ 1.2 GB; total > 24 TB.
Revised color profiles applied to 300+ slides (see
revised_list.csv
).
4. Evaluation Focus
Clinical accuracy over BLEU / ROUGE.
Real-world diagnostic value ranked by expert pathologists.
Bonus for concise, well-structured sections (macro / micro findings, impressions).
5. Recommended Workflow
Patch extraction + H&E normalization
ViT-L or SAM-ViT encoder fine-tuned on color-corrected slides
Seq-to-Seq decoder conditioned on slide embeddings + prior draft text
Rule-based medical NER layer to normalise terminology
Continuous evaluation with the provided GitHub script
6. Revised Schedule (KST 10 AM)
MilestoneOldNew (final)Training data release13 May 202520 May 2025Registration deadline27 Jun 202504 Jul 2025Debug Phase27 Jun → 03 Jul04 Jul → 10 JulTest 1 opens11 Jul 2025unchangedTest 1 deadline18 Jul → 25 Jul01 Aug 2025Test 2 opens02 Aug 2025unchangedTest 2 deadline09 Aug → 16 Aug23 Aug 2025Winners announced—09 Sep 2025
7. Quick Links
Registration & rules:
https://reg2025.grand-challenge.org/
Training dataset mirrors: Reg2025-TrainDataset page
Evaluation toolkit: GitHub → Submission & Evaluation
8. Final Checklist
Download revised TIFFs or run
tiffset
locally.Validate output JSON in Debug Phase.
Monitor further notices in the Updates channel.


© 2024 Pathfinder Lab. All rights reserved.