search

Tencent / tencent-ml-images

Largest multi-label image database; ResNet-101 model; 80.73% top-1 acc on ImageNet
Other
3.05k stars 514 forks source link

请问是否可以在不下载数据进行训练的情况下，进行demo测试？ #7

Open btdan opened 5 years ago

btdan commented 5 years ago

如题，请问是否可以不直接下载数据进行训练，而是通过您网页中提供的Checkpoints ckpt-resnet101-mlimages ckpt-resnet101-mlimages-imagenet
来进行demo测试？是否辛苦提供一下具体步骤？非常感谢。 Thanks very much

wubaoyuan commented 5 years ago

@btdan Maybe you can check './example/finetune.sh'? Just replace the ImageNet images with your own images, and generate the tfrecord file using './tfrecord.sh'

btdan commented 5 years ago

非常感谢您的回复。我的意思是，是否有现成的模型来直接对某张图像进行测试。我主要是想了解一下对于小目标物体的识别效果？恳请您能否告诉我具体的操作步骤？非常感谢。 Thanks very much.

wubaoyuan commented 5 years ago

@btdan I get it. We will add such a demo soon. But you can firstly check the similar demo is finetune.sh, which loads the checkpoint, and extract the features of the image.

wubaoyuan commented 5 years ago

@btdan It is a good suggestion :)

btdan commented 5 years ago

非常感谢，很期待您的更新。多谢

wubaoyuan commented 5 years ago

@btdan the demo for image classification has been added. Please check the updated README

Jason-xin commented 5 years ago

@wubaoyuan The demo is based on imagenet dictionary, can you provide a demo based on ML-Image dictionary?? You have provided two pretrained model, are both trained on ImageNet dictionary?? instead of ML-Image dictionary? image

wubaoyuan commented 5 years ago

@Jason-xin Hi, the first ckpt is pre-trained with the ML-Images dictionary. You suggestion is good, we plan to add a demo to use this ckpt.

Jason-xin commented 5 years ago

@wubaoyuan Thanks，maybe it just need add ML-Images dictionary text file? If not, hope for your new demo in ML-Images soon.

wubaoyuan commented 5 years ago

@Jason-xin Yes. As the output layers of ResNet for both ML-Images and ImageNet are M (size of dictionary) sigmoid functions, one just need to replace the dictionary.

Jason-xin commented 5 years ago

@wubaoyuan Thank you! I have done~ Hope for your paper ^-^

xxllp commented 5 years ago

@Jason-xin ，have you try it out?

Easyfeng222 commented 5 years ago

@wubaoyuan Excuse me, When can I open the test code for ML-images? Looking forward to your reply

Easyfeng222 commented 5 years ago

@Jason-xin Excuse me, have you completed the ML-images test? If you can, can you talk about the specific method? Thank you in advance.

Jason-xin commented 5 years ago

@Easyfeng222 I just test the single-label classification on ML-Images, you can just change "--dictionary" param in /example/image_classification.sh.