---
title: pvt_v2_b3
canonical_url: "https://www.modelscope.cn/models/Shanghai_AI_Laboratory/pvt_v2_b3"
md_url: "https://www.modelscope.cn/models/Shanghai_AI_Laboratory/pvt_v2_b3.md"
repository: Shanghai_AI_Laboratory/pvt_v2_b3
last_updated: 2024-12-30
license: apache-2.0
pipeline_tag: image-classification
tasks:
  - image-classification
model_type:
  - pvt_v2
architectures:
  - PvtV2ForImageClassification
parameters: 45.2M
tensor_type:
  - F32
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 1103
stars: 0
---

# pvt_v2_b3

> pvt_v2_b3 - Shanghai_AI_Laboratory 在 ModelScope 开源的模型。This is the Hugging Face PyTorch implementation of the PVTv2 model.

Shanghai_AI_Laboratory/pvt_v2_b3 是 ModelScope 魔搭社区上的 45.2M 参数image-classification模型，采用 apache-2.0 许可。

- **Repository**: Shanghai_AI_Laboratory/pvt_v2_b3
- **License**: apache-2.0
- **Tasks**: image-classification
- **Parameters**: 45.2M
- **Downloads**: 1103
- **Stars**: 0
- **Last updated**: 2024-12-30

Source: https://www.modelscope.cn/models/Shanghai_AI_Laboratory/pvt_v2_b3

---

# PVTv2

This is the Hugging Face PyTorch implementation of the [PVTv2](https://arxiv.org/abs/2106.13797) model.

## Model Description

The Pyramid Vision Transformer v2 (PVTv2) is a powerful, lightweight hierarchical transformer backbone for vision tasks. PVTv2 infuses convolution operations into its transformer layers to infuse properties of CNNs that enable them to learn image data efficiently. This mix transformer architecture requires no added positional embeddings, and produces multi-scale feature maps which are known to be beneficial for dense and fine-grained prediction tasks.

Vision models using PVTv2 for a backbone:
1. [Segformer](https://arxiv.org/abs/2105.15203) for Semantic Segmentation.
2. [GLPN](https://arxiv.org/abs/2201.07436) for Monocular Depth.
3. [Deformable DETR](https://arxiv.org/abs/2010.04159) for 2D Object Detection.
4. [Panoptic Segformer](https://arxiv.org/abs/2109.03814) for Panoptic Segmentation.