Mask2former Swin Base IN21k Coco Instance

🚀 Mask2Former

Mask2Former是一个用于图像分割的模型，它能以相同的范式处理实例、语义和全景分割任务。该模型在COCO实例分割任务上进行了训练，具有出色的性能和效率。

🚀 快速开始

Mask2Former模型基于特定范式处理实例、语义和全景分割任务，即将所有任务都当作实例分割来处理。它在性能和效率上超越了之前的SOTA模型MaskFormer。以下是使用该模型的示例代码：

import requests
import torch
from PIL import Image
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation


# load Mask2Former fine-tuned on COCO instance segmentation
processor = AutoImageProcessor.from_pretrained("facebook/mask2former-swin-base-IN21k-coco-instance")
model = Mask2FormerForUniversalSegmentation.from_pretrained("facebook/mask2former-swin-base-IN21k-coco-instance")

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)

# model predicts class_queries_logits of shape `(batch_size, num_queries)`
# and masks_queries_logits of shape `(batch_size, num_queries, height, width)`
class_queries_logits = outputs.class_queries_logits
masks_queries_logits = outputs.masks_queries_logits

# you can pass them to processor for postprocessing
result = processor.post_process_instance_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
# we refer to the demo notebooks for visualization (see "Resources" section in the Mask2Former docs)
predicted_instance_map = result["segmentation"]

更多代码示例请参考文档。

✨ 主要特性

• 统一范式：Mask2Former使用相同的范式处理实例、语义和全景分割任务，将所有任务都视为实例分割。
• 性能优越：通过采用更先进的多尺度可变形注意力Transformer替换像素解码器、使用带掩码注意力的Transformer解码器以及在子采样点上计算损失等方法，在性能和效率上超越了之前的SOTA模型MaskFormer。

📚 详细文档

模型描述

Mask2Former通过预测一组掩码和相应的标签来处理实例、语义和全景分割任务。它将所有3个任务都当作实例分割来处理，在性能和效率上超越了之前的SOTA模型MaskFormer，具体改进如下：

• 用更先进的多尺度可变形注意力Transformer替换像素解码器。
• 采用带掩码注意力的Transformer解码器，在不引入额外计算的情况下提升性能。
• 通过在子采样点上计算损失而不是在整个掩码上计算，提高训练效率。

预期用途与限制

你可以使用此特定检查点进行实例分割。可查看模型中心以查找针对你感兴趣的任务进行微调的其他版本。

使用说明

以下是使用该模型的示例代码：

import requests
import torch
from PIL import Image
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation


# load Mask2Former fine-tuned on COCO instance segmentation
processor = AutoImageProcessor.from_pretrained("facebook/mask2former-swin-base-IN21k-coco-instance")
model = Mask2FormerForUniversalSegmentation.from_pretrained("facebook/mask2former-swin-base-IN21k-coco-instance")

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)

# model predicts class_queries_logits of shape `(batch_size, num_queries)`
# and masks_queries_logits of shape `(batch_size, num_queries, height, width)`
class_queries_logits = outputs.class_queries_logits
masks_queries_logits = outputs.masks_queries_logits

# you can pass them to processor for postprocessing
result = processor.post_process_instance_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
# we refer to the demo notebooks for visualization (see "Resources" section in the Mask2Former docs)
predicted_instance_map = result["segmentation"]

📄 许可证

许可证类型：其他

信息表格

免责声明

发布Mask2Former的团队未为此模型编写模型卡片，此模型卡片由Hugging Face团队编写。