Mask2Former模型在COCO实例分割上进行了训练(基于尺寸的版本,使用Swin骨干)。该模型在 Masked-attention Mask Transformer for Universal Image Segmentation 论文中首次提出,并于 this repository 首次发布。
免责声明:发布Mask2Former的团队没有为该模型编写模型卡片,所以此模型卡片由Hugging Face团队编写。
Mask2Former通过预测一组掩码和相应的标签来解决实例分割、语义分割和全景分割的问题。因此,这三个任务都被视为实例分割。Mask2Former通过以下方式在性能和效率上优于之前的SOTA模型 MaskFormer :(i) 用更先进的多尺度可变形注意力Transformer替代像素解码器,(ii) 采用具有掩码注意力的Transformer解码器,在不引入额外计算的同时提高性能,(iii) 通过在子采样点上计算损失而不是整个掩码来提高训练效率。
您可以使用此特定检查点进行实例分割。请参阅 model hub 以查找您感兴趣的任务的其他微调版本。
下面是如何使用此模型的方法:
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-coco-instance") model = Mask2FormerForUniversalSegmentation.from_pretrained("facebook/mask2former-swin-base-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"]
有关更多代码示例,请参考 documentation 。