1 Star 0 Fork 455

SelwynLee / mindformers_r1.0

forked from MindSpore / mindformers 
加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
克隆/下载
zero_shot_image_classification.md 4.01 KB
一键复制 编辑 原始数据 按行查看 历史
Lin 提交于 2023-11-08 16:01 . fix docs.

Zero Shot Image Classification

任务描述

零样本图像分类:模型在基于图文对的预训练后,可以在给定任意图片与候选标签列表的情况下,完成对图像的分类,而无需任何微调。

相关论文 Alec Radford, Jong Wook Kim, et al., Learning Transferable Visual Models From Natural Language Supervision, 2021.

已支持数据集性能

model type datasets Top1-accuracy stage example
clip clip_vit_b_32
clip_vit_b_16
clip_vit_l_14
clip_vit_l_14@336
Cifar100 57.24%
61.41%
69.67%
68.19%
eval
predict
link
link

Cifar100

  • 数据集大小:161M,共60000张图片,100个类别
    • 训练集:50000张图片
    • 测试集:10000张图片
  • 数据格式:二进制文件
数据集目录格式
└─cifar-100-python
   ├─meta
   ├─test  
   └─train  

快速任务接口

  • Trainer接口开启评估/推理:
import mindspore; mindspore.set_context(mode=0, device_id=0)
from mindformers import MindFormerBook
from mindformers.trainer import Trainer
from mindformers.tools.image_tools import load_image

# 显示Trainer的模型支持列表
MindFormerBook.show_trainer_support_model_list("zero_shot_image_classification")
# INFO - Trainer support model list for zero_shot_image_classification task is:
# INFO -    ['clip_vit_b_32', 'clip_vit_b_16', 'clip_vit_l_14', 'clip_vit_l_14@336']
# INFO - -------------------------------------

# 初始化trainer
trainer = Trainer(task='zero_shot_image_classification',
    model='clip_vit_b_32',
    eval_dataset='cifar-100-python'
)
img = load_image("https://ascend-repo-modelzoo.obs.cn-east-2."
          "myhuaweicloud.com/XFormer_for_mindspore/clip/sunflower.png")
trainer.evaluate()  #下载权重进行评估
# INFO - Top1 Accuracy=57.24%
trainer.predict(input_data=img)  #下载权重进行推理
# INFO - output result is saved at ./results.txt
  • pipeline接口开启快速推理
import mindspore; mindspore.set_context(mode=0, device_id=0)
from mindformers import pipeline, MindFormerBook
from mindformers.tools.image_tools import load_image

# 显示pipeline支持的模型列表
MindFormerBook.show_pipeline_support_model_list("zero_shot_image_classification")
# INFO - Pipeline support model list for zero_shot_image_classification task is:
# INFO -    ['clip_vit_b_32', 'clip_vit_b_16', 'clip_vit_l_14', 'clip_vit_l_14@336']
# INFO - -------------------------------------

# pipeline初始化
classifier = pipeline("zero_shot_image_classification",
                      model="clip_vit_b_32"
                      candidate_labels=["sunflower", "tree", "dog", "cat", "toy"])
img = load_image("https://ascend-repo-modelzoo.obs.cn-east-2."
          "myhuaweicloud.com/XFormer_for_mindspore/clip/sunflower.png")
classifier(img)
# 输出
# [[{'score': 0.99995565, 'label': 'sunflower'}, {'score': 2.5318595e-05, 'label': 'toy'},
# {'score': 9.903885e-06, 'label': 'dog'}, {'score': 6.75336e-06, 'label': 'tree'},
# {'score': 2.396818e-06, 'label': 'cat'}]]
Python
1
https://gitee.com/selwynlee/mindformers_1.git
git@gitee.com:selwynlee/mindformers_1.git
selwynlee
mindformers_1
mindformers_r1.0
dev

搜索帮助