| Property | Value |
|---|---|
| Parameter Count | 213M |
| Top-1 Accuracy | 86.52% |
| License | Apache 2.0 |
| Paper | MaxViT: Multi-Axis Vision Transformer |
| Input Resolution | 512x512 |
What is maxvit_large_tf_512.in1k?
MaxViT Large is a sophisticated vision transformer model that combines the strengths of convolution operations with multi-axis attention mechanisms. Originally trained in TensorFlow and ported to PyTorch, this model represents a significant advancement in vision transformer architecture, designed to process high-resolution images at 512x512 pixels.
Implementation Details
The model implements a hybrid architecture that utilizes both MBConv (mobile inverted bottleneck) blocks and dual-path attention mechanisms. With 212.33M parameters and 244.75 GMACs, it offers a balance between computational efficiency and model performance.
- Combines convolutional blocks with window and grid attention mechanisms
- Features a large-scale architecture optimized for 512x512 input resolution
- Implements LayerNorm for normalization throughout the network
- Achieves 86.52% top-1 accuracy on ImageNet-1K dataset
Core Capabilities
- High-resolution image classification with state-of-the-art performance
- Feature extraction for downstream computer vision tasks
- Efficient processing of large images through multi-axis attention
- Balanced trade-off between computational cost and accuracy
Frequently Asked Questions
Q: What makes this model unique?
This model introduces a novel multi-axis attention mechanism that processes visual information across different spatial partitioning schemes, combining the benefits of both local and global attention patterns with convolutional operations.
Q: What are the recommended use cases?
The model is particularly well-suited for high-resolution image classification tasks, computer vision applications requiring detailed feature extraction, and scenarios where processing larger images is necessary while maintaining high accuracy.