hikapok / tf.fashionai Goto Github PK

您好 我最近在做服装属性的分类研究 希望您能发送一份数据集给我[email protected] 感激不尽 谢谢 官网迟迟不公布 很急 谢谢

How to test the pre-trained models in my own images?

How to train your code with multiply gpus? THANKS

First of all, thanks for the sharing.
I took part in this competition as well and I also used CPN. However, I got some findings different with yours during my experiments. Hope we can exchange some ideas.

DetNet is better, perform almost the same as SEResNeXt, while SEResNet showed little improvement than ResNet

I also tried DetNet as backbone net but got a bad result. I guess it was because I trained it from scratch. One interesting is that, in my work, ResNet152 outperforms other backbone nets including ResNet-InceptionV2, SENet and NasNet.

Enforce the loss of invisible keypoints to zero gave better performance

I tried both but didn't find a great difference here. So is there a significant improvement in your case?

It's bad to do gaussian blur on the predicted heatmap

Still, no big difference for me, did you find it very worse?

Ensemble of the heatmaps for fliped images is worser than emsemble of the predictions of fliped images

I got an opposite result for this. Maybe it depends on the model.

Can you send me the dataset of key points for fashionai clothing? The download url have closed,Email:[email protected]. thanks

Hello, looking at this experiment is very good,and if there are more than one person in a picture, Does the model work well?

Hello,thanks for your great work. I want to know how to put all catgories into one model rather than training separate ones . Could you please give me some suggestions in detail? I am looking forward to your reply,thank you very much.

Hello
I just tried to download datasets from the page of contest you provided in readme, but I'm not able to download data from here(
Don't you have this data stored in some other place?
Thank you in advance