Research map / Image generation / Image restoration
CHAMMI-75: pre-training multi-channel models with heterogeneous microscopy images
Vidit Agrawal, John Peters, Tyler N. Thompson, Mohammad Vali Sanian, Chau Pham, Nikita Moshkov, Arshad Kazi, Aditya Pillai, Jack Freeman, Byunguk Kang, Samouil L. Farhi, Ernest Fraenkel and 4 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on CHAMMI-75 (DINO-BoC) instead of IDRCell100k | DINO-BoC · CHAMMI | benchmark score: 37.38 -> 48.75 | Improved |
| pre-train on CHAMMI-75 instead of target data only (ablation) | DINO-BoC · CHAMMI scores | relative performance on CHAMMI scores: 27% -> 38% | Improved |
| train on only the two dominant modalities instead of other 12 modalities (ablation) | DINO-BoC · CHAMMI scores | relative performance on CHAMMI scores: 13% -> 15% | Hurt |
| vary microscope magnification diversity (ablation) | DINO-BoC · CHAMMI scores | relative performance on CHAMMI scores: ~3% | Improved |
| remove most common cell lines U2OS and A549 (ablation) | DINO-BoC · CHAMMI scores | relative performance on CHAMMI scores: ~1% | No clear effect |
| limit studies to up to two, three or four channels (ablation) | DINO-BoC · CHAMMI scores | relative performance on CHAMMI scores: ~1% | No clear effect |
| bag of channels instead of multi-channel attention under SSL (ablation) | DINO · CHAMMI scores | relative performance on CHAMMI scores: up to 19% relative | Improved |
| use DINO instead of MAE (ablation) | BoC models · CHAMMI scores | relative performance on CHAMMI scores: ~15% relative | Improved |
| use DINO instead of SimCLR (ablation) | BoC models · CHAMMI scores | relative performance on CHAMMI scores: ~7% relative | Improved |
| scale model from ViT-small to ViT-large (ablation) | BoC SSL · CHAMMI scores | relative performance on CHAMMI scores: 10% relative | Improved |
| add weak supervision with noisy labels (ablation) | MCA-SSL · CHAMMI scores | relative performance on CHAMMI scores: 1-19% relative | Improved |
| train on full 5x larger CHAMMI-75 (ablation) | MorphEm BoC ViT-small DINO · CHAMMI scores | relative performance on CHAMMI scores: 9.8% relative | Improved |
Closest papers by meaning
- MorphoHELM: A Comprehensive Benchmark for Evaluating Representations for Microscopy-Based Morphology Assays (NeurIPS 2026 · similarity 0.88)
- ChAda-ViT : Channel Adaptive Attention for Joint Representation Learning of Heterogeneous Microscopy Images (CVPR 2024 · similarity 0.87)
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.86)
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.85)
- Semantic and Visual Crop-Guided Diffusion Models for Heterogeneous Tissue Synthesis in Histopathology (NeurIPS 2025 · similarity 0.85)
- CellCLIP - Learning Perturbation Effects in Cell Painting via Text-Guided Contrastive Learning (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
