File size: 2,583 Bytes
a980711 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 |
---
language:
- en
tags:
- computer-vision
- segmentation
- few-shot-learning
- zero-shot-learning
- sam2
- clip
- pytorch
license: apache-2.0
datasets:
- custom
metrics:
- iou
- dice
- precision
- recall
library_name: pytorch
pipeline_tag: image-segmentation
---
# Model Card for SAM 2 Few-Shot/Zero-Shot Segmentation
## Model Description
This repository contains two main models for domain-adaptive segmentation:
### SAM2FewShot
- **Architecture**: SAM 2 + CLIP with memory bank
- **Purpose**: Few-shot learning for segmentation
- **Input**: Images + support examples
- **Output**: Segmentation masks
### SAM2ZeroShot
- **Architecture**: SAM 2 + CLIP with advanced prompting
- **Purpose**: Zero-shot learning for segmentation
- **Input**: Images + text prompts
- **Output**: Segmentation masks
## Intended Uses & Limitations
### Primary Use Cases
- Domain adaptation for segmentation tasks
- Rapid deployment in new environments
- Minimal supervision scenarios
- Research in few-shot/zero-shot learning
### Limitations
- Performance depends on prompt quality
- Domain-specific adaptations required
- Computational cost of attention mechanisms
- Limited cross-domain generalization
## Training and Evaluation Data
### Domains
- **Satellite Imagery**: Buildings, roads, vegetation, water
- **Fashion**: Shirts, pants, dresses, shoes
- **Robotics**: Robots, tools, safety equipment
### Evaluation Metrics
- IoU (Intersection over Union)
- Dice coefficient
- Precision and Recall
- Boundary accuracy
- Hausdorff distance
## Training Results
### Few-Shot Performance (5 shots)
| Domain | Mean IoU | Mean Dice |
|--------|----------|-----------|
| Satellite | 65% | 71% |
| Fashion | 62% | 68% |
| Robotics | 59% | 65% |
### Zero-Shot Performance (Best Strategy)
| Domain | Mean IoU | Mean Dice |
|--------|----------|-----------|
| Satellite | 42% | 48% |
| Fashion | 38% | 45% |
| Robotics | 35% | 42% |
## Environmental Impact
- **Hardware Type**: GPU (NVIDIA V100 recommended)
- **Hours used**: Variable based on experiments
- **Cloud Provider**: Any cloud with GPU support
- **Compute Region**: Any
- **Carbon Emitted**: Depends on usage
## Citation
```bibtex
@misc{sam2_fewshot_zeroshot_2024,
title={SAM 2 Few-Shot/Zero-Shot Segmentation: Domain Adaptation with Minimal Supervision},
author={Your Name},
year={2024},
url={https://huggingface.co/esalguero/Segmentation}
}
```
## Model Card Authors
This model card was written by the research team.
## Model Card Contact
For questions about this model card, please contact the repository maintainers. |