Segformer B3 1024x1024 City 160k

🚀 Segformer模型卡片

Segformer是一个用于图像分割的模型，借助该模型可以实现对图像中不同对象的精准分割，在计算机视觉领域有着广泛的应用。

🚀 快速开始

加载预训练模型

点击下面的按钮在Colab中打开示例：

安装依赖

pip install -U segmentation_models_pytorch albumentations

运行推理

import torch
import requests
import numpy as np
import albumentations as A
import segmentation_models_pytorch as smp

from PIL import Image

device = "cuda" if torch.cuda.is_available() else "cpu"

# 加载预训练模型和预处理函数
checkpoint = "smp-hub/segformer-b3-1024x1024-city-160k"
model = smp.from_pretrained(checkpoint).eval().to(device)
preprocessing = A.Compose.from_pretrained(checkpoint)

# 加载图像
url = "https://huggingface.co/datasets/hf-internal-testing/fixtures_ade20k/resolve/main/ADE_val_00000001.jpg"
image = Image.open(requests.get(url, stream=True).raw)

# 预处理图像
np_image = np.array(image)
normalized_image = preprocessing(image=np_image)["image"]
input_tensor = torch.as_tensor(normalized_image)
input_tensor = input_tensor.permute(2, 0, 1).unsqueeze(0)  # HWC -> BCHW
input_tensor = input_tensor.to(device)

# 进行推理
with torch.no_grad():
    output_mask = model(input_tensor)

# 后处理掩码
mask = torch.nn.functional.interpolate(
    output_mask, size=(image.height, image.width), mode="bilinear", align_corners=False
)
mask = mask.argmax(1).cpu().numpy()  # 在预测的类别（通道维度）上取最大值

📦 安装指南

安装必要的依赖库，命令如下：

pip install -U segmentation_models_pytorch albumentations

💻 使用示例

基础用法

import torch
import requests
import numpy as np
import albumentations as A
import segmentation_models_pytorch as smp

from PIL import Image

device = "cuda" if torch.cuda.is_available() else "cpu"

# 加载预训练模型和预处理函数
checkpoint = "smp-hub/segformer-b3-1024x1024-city-160k"
model = smp.from_pretrained(checkpoint).eval().to(device)
preprocessing = A.Compose.from_pretrained(checkpoint)

# 加载图像
url = "https://huggingface.co/datasets/hf-internal-testing/fixtures_ade20k/resolve/main/ADE_val_00000001.jpg"
image = Image.open(requests.get(url, stream=True).raw)

# 预处理图像
np_image = np.array(image)
normalized_image = preprocessing(image=np_image)["image"]
input_tensor = torch.as_tensor(normalized_image)
input_tensor = input_tensor.permute(2, 0, 1).unsqueeze(0)  # HWC -> BCHW
input_tensor = input_tensor.to(device)

# 进行推理
with torch.no_grad():
    output_mask = model(input_tensor)

# 后处理掩码
mask = torch.nn.functional.interpolate(
    output_mask, size=(image.height, image.width), mode="bilinear", align_corners=False
)
mask = mask.argmax(1).cpu().numpy()  # 在预测的类别（通道维度）上取最大值

🔧 技术细节

模型初始化参数

model_init_params = {
    "encoder_name": "mit_b3",
    "encoder_depth": 5,
    "encoder_weights": None,
    "decoder_segmentation_channels": 768,
    "in_channels": 3,
    "classes": 19,
    "activation": None,
    "aux_params": None
}

数据集

数据集名称：Cityscapes

📚 详细文档

• 库：https://github.com/qubvel/segmentation_models.pytorch
• 文档：https://smp.readthedocs.io/en/latest/

📄 许可证

许可证信息：https://github.com/NVlabs/SegFormer/blob/master/LICENSE

本模型已使用 PytorchModelHubMixin 推送到模型中心。