youngkyunjang / gpq Goto Github PK
View Code? Open in Web Editor NEWGeneralized Product Quantization Network For Semi-supervised Image Retrieval - CVPR 2020
License: MIT License
Generalized Product Quantization Network For Semi-supervised Image Retrieval - CVPR 2020
License: MIT License
hello , I can not download pretrained model , can you fix it?
Unfortunately, my server cannot run VGG network because there are too many network parameters. So I want to use CNN-F (Alexnet) instead of VGG. Two Global_Average_Pooling methods are used in the VGG network and the full connection layer is removed. Can you tell me how to modify in CNN-F(AlexNet) network?
thanks!
It is said, "S_b in the N-PQ loss (i.e., Equation 2 in the paper) denotes a cosine similarity between b-th feature vector and every quantized vector". According to the code, this is implemented by matrix multiplication:
Line 99 in 8096383
embeddings_x
(i.e., feature_S) is intra-normalized, but embeddings_q
(i.e., descriptor_S) is NOT. (This can be verified by printing the sub-vector norm.)
So, this is actually not the cosine similarity, but just an ordinary inner product. The product does not reflect a similarity measure in the conventional sense, and may affect the results.
1.hello, what are the specific mean and variance in nuswide?
def color_preprocessing(x_):
#Normalize with mean and std of source data
x_ = x_.astype('float32')
x_[:, :, :, 0] = (x_[:, :, :, 0] - 125.642) / 63.01
x_[:, :, :, 1] = (x_[:, :, :, 1] - 123.738) / 62.157
x_[:, :, :, 2] = (x_[:, :, :, 2] - 114.46) / 66.94
return x_
Good work and thanks!
If you have the code of the pytorch version, please share it!
Nice work and Thanks!
Meanwhile the downloaded mat for CIFAR10 can't be loaded correctly.
Hi,
There are a total of 27000 images in the NUS dataset. I chose the 21 most popular classes, with a total of 195834 images. But, the experimental results are much worse than those in the paper. Can you provide the name list and label of train and test image database.
The experimental results obtained by using Alexnet or VGG network on experimental protocol 2 ?
I find the program is computing distance based on table Z.
but this way is hard to applying my own experiments since I want to extract the hash codes for my custom dataset. Then I can plot some precision recall curve and so on.
could you add the way of extracting hash codes for some dataset?
Hello, I was that in the issues the problem about pretrained model is mentioned and then closed by fixing the issue. But the link for the pretrained model weights is still broken. Thanks
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.