personal projects
Neural Morphological Analysis of Fungi
A microscopy project for fungal species classification and future neural-field modeling of morphology from high-resolution biological image patches.
---
project status: alpha stage
---
Current Status: Phase 1 (Classification)
We are currently in the initial phase of the project, establishing a robust classification baseline for 9 different fungal species.
- Dataset Integrated: High-resolution microscopic images (3600x5760) from the Fungus Dataset.
- Patch-Based Learning: Implementation of a ResNet18-based classifier that learns from 224x224 patches, guided by biological masks to focus specifically on fungal structures.
- Explainability: Integrated Grad-CAM (Gradient-weighted Class Activation Mapping) to visualize and interpret the features the model uses for identification.
Next Steps: Neural Morphological Signatures
Once the classification baseline is stabilized, we will move towards:
- Neural Fields (SIREN): Learning implicit representations of fungal morphology for infinite-resolution analysis.
- Latent Morphing: Exploring the morphological transitions between species in a shared latent space.
- VLM Grounding: Using Vision-Language Models (like CLIP) for open-vocabulary querying of fungal features.