akarshzingade

Well, the weight file linked in this repository is 275MB. I am not sure about the orientation part.

Maybe you can use different architectures for YOLOv2- like tinyYOLO. It has a lower memory footprint. Once you get the bounding box for your logo, you can use orientation detection...

Hey Mike and Longzeyilang! Sub-sampling is same as strides. "subsample: tuple of length 2. Factor by which to subsample output. Also called strides elsewhere."- https://faroit.github.io/keras-docs/1.2.2/layers/convolutional/#convolution2d

Hey, Longzeyilang. I believe the implementation does follow the architecture shown in Figure 3. Please let me know what the difference is :)

Hey Longzeyilang! 1) Do you mean the parallel smaller network with VGG? If yes, I do have implemented it :) 2) The pairwise relevance score is hand crafted. Unfortunately, they...

Yes, that is correct! I was supposed to change this. But, somehow forgot about it! Thank you for pointing that out :)

So, the model is relying on the color too much?

I think Inception is less sensitive to colour. You could try that. I haven't found any article/paper that shows the colour sensitivity of VGG. The closest I have found is...

50 per class is fine I think. It's the number of triplets per query image that matters. I would say 50 triplets per query image and positive image pair.

using "--num_pos_images 10 --num_neg_images 40" will create 40*10 triplets per query. What I meant to say was 50 negative images per query image and positive image pair. But, this will...