FashionAI icon indicating copy to clipboard operation
FashionAI copied to clipboard

How to organize data sets?

Open Kiritooo opened this issue 6 years ago • 8 comments

hello, I'm new of the CV, I want to use your code to understand how to use dataset for pose estimation, but I don't know how to deal with data sets and name, I will not submit results to match only for learning, can you give a README thank you!

Kiritooo avatar Mar 13 '18 13:03 Kiritooo

@Kiritooo We add README and will gradually improve it. If you want to know the approach of deal with data sets, please read datasets.py carefully.

Lmy0217 avatar Mar 14 '18 05:03 Lmy0217

兄弟,线上能到多少?

wuhaotju avatar Mar 21 '18 12:03 wuhaotju

I find that the Accuracy is very low, what about you?

shiontao avatar Mar 21 '18 16:03 shiontao

Train Epoch: 16 [7680/9056 (85%)] Loss: 1.734629 Train Epoch: 16 [8320/9056 (92%)] Loss: 1.775722 Train Epoch: 16 [8960/9056 (99%)] Loss: 1.782988 Test set: Average loss: 2.9240, Accuracy: 369/2264 (16%) Train Epoch: 17 [0/9056 (0%)] Loss: 1.722289 Train Epoch: 17 [640/9056 (7%)] Loss: 1.780630 Train Epoch: 17 [1280/9056 (14%)] Loss: 1.787481 Train Epoch: 17 [1920/9056 (21%)] Loss: 1.722810 Train Epoch: 17 [2560/9056 (28%)] Loss: 1.876558 Is that normal?

shiontao avatar Mar 21 '18 16:03 shiontao

@rmfckaka I think it may be caused by overfitting, try image enhancement algorithm to enlarge your datasets. And the train loss increased , I guess u may take a smaller lr to train the model.

augusyan avatar Mar 22 '18 04:03 augusyan

@rmfckaka 我没有尝试

wuhaotju avatar Mar 23 '18 01:03 wuhaotju

分数百分之几?

aigeek avatar Apr 20 '18 05:04 aigeek

我的准确率也不高,没参加比赛,想试一下这个任务,有加群交流的吗(102611529)

wzjiang avatar May 15 '18 03:05 wzjiang