psinger/kaggle-landmark-recognition-2020-1st-place

kaggle-landmark-recognition-2020-1st-place's Issues

s and m in ArcFace loss

Thanks for sharing your great work.
I noticed that you have s=45, m=0.1 by default in the loss function. In your configx.py, you have s=45, m=0.3,0.4 etc. In the original ArcFace paper, the default s=60, m=0.5. Did you optimize the s and m in your work?

why did you only use small data for train not all train data

first of all. thanks for your cool work. why not use all train data to train model ?

kaggle-landmark-recognition-2020-1st-place/src/train.py

Line 93 in a4a4671

train = train.head(args.batch_size*2)

what is the train dataset in your post-processing ranking?

Your paper says,

Let Test refer to the images from the test set to be recognized and ranked, Train to the set of candidate images restricted to all possible landmarks from the test set, and Non-landmark as an additional set of out�of-domain non-landmark images

What does exactly it mean by "Train to the set of candidate images restricted to all possible landmarks from the test set"? Is there an annotation which images in the test set are landmarks?

Thanks.

Hi

I am very interested in your project, is there a pre-trained model to test