Frequently Asked Questions (FAQ)

General Guidelines

This FAQ addresses common questions about participation requirements, data usage policies, and technical specifications for the SLC-PFM competition. For additional inquiries not covered here, please contact slcpfm2025@gmail.com.

🔧 Technical Questions

Q1: Are participants permitted to fine-tune existing foundation models, or must models be trained entirely from scratch?

Answer: Participants are permitted to use publicly available pathology foundation models as starting points for their submissions. However, only foundation models that are publicly accessible and documented in this foundation models list may be utilized.

Important restrictions:

• Private or proprietary pathology foundation models are strictly prohibited
• Models trained on internal institutional datasets cannot be used
• Only models with public access and documentation are acceptable

Q2: Can existing foundation models (e.g., UNI, PLIP) be incorporated as guidance during the pre-training phase?

Answer: Yes, participants may leverage publicly available foundation models to inform or assist portions of their self-supervised learning methodology, provided the following conditions are met:

• The foundation models must be publicly accessible
• Models must be included in this foundation models list
• Proper attribution and methodology description must be provided in the submission documentation

Q3: Are participants allowed to incorporate additional information derived from the competition images using external models?

Answer: Yes, participants may generate supplementary information from the competition dataset using publicly available models, subject to the following guidelines:

Permitted approaches:

• Image descriptions generated by public large language models (LLMs)
• Feature extraction using publicly available vision models
• Metadata generation using open-source tools

Requirements:

• All models used for information generation must be publicly available
• Proper citation and attribution required in submission documentation
• Generated information must be derived solely from the competition dataset

Restrictions:

• External or private datasets cannot be used to enhance generative models
• Proprietary models trained on private pathology data are prohibited

Q4: Can participants access de-identified whole slide images (WSIs) to obtain spatial location information?

Answer: No, raw whole slide images cannot be provided due to institutional data governance policies. The competition dataset consists exclusively of pre-processed image patches extracted from whole slide images.

Available data format:

• Pre-cropped pathology image tiles
• Standardized resolution and format
• No spatial coordinate information included

Rationale: This limitation ensures compliance with institutional review board requirements and patient privacy protections while maintaining dataset consistency across all participants.

📋 Dataset and Usage Policies

Q5: What external data sources are permitted for this competition?

Answer: Participants must use only the provided competition dataset for model training and validation. External pathology datasets, whether public or private, are strictly prohibited to ensure fair comparison across all submissions.

Permitted external resources:

• Publicly available pre-trained models (from the available list)
• Open-source software libraries and frameworks
• Public generative models for information extraction from competition data

Q6: Are there restrictions on computational resources or training time?

Answer: While there are no explicit limits on computational resources for model training, submitted models must meet the specified inference requirements:

• Single GPU deployment capability
• Maximum 80GB GPU memory usage during inference
• Efficient feature extraction for downstream evaluation

📝 Submission Requirements

Q8: What documentation is required for foundation model submissions?

Answer: Each submission must include comprehensive technical documentation covering:

• Methodology description: Detailed explanation of self-supervised learning approach
• Architecture specifications: Model design and implementation details
• Training procedures: Hyperparameters, computational requirements, training duration
• External resource attribution: Proper citation of any publicly available models or tools used
• Performance validation: Any internal evaluation or ablation studies conducted

📧 Contact Information

For questions not addressed in this FAQ, please contact our organizing committee:

Email: slcpfm2025@gmail.com
Subject Line Format: “SLC-PFM Question: [Topic]”

Competition Website: https://deeplearningpathology.org/

This FAQ is updated regularly based on participant inquiries. Please check back periodically for new information and clarifications.