naver / cgd Goto Github PK
View Code? Open in Web Editor NEWCombination of Multiple Global Descriptors for Image Retrieval
Home Page: https://arxiv.org/abs/1903.10663
License: Apache License 2.0
Combination of Multiple Global Descriptors for Image Retrieval
Home Page: https://arxiv.org/abs/1903.10663
License: Apache License 2.0
You have done a very nice job on your paper! I tried to implement your proposed network these days, but I found several problems.
The first one is I found that the Fc layer after descriptor always have negative impacts on the result, on the CUB200-2011, I get top-1 recall 72 on the final l2 layer, but I can get 76 on GD(1) layer (under MG configuration). I think maybe it is influenced strongly by the auxiliary classification branch? I want to know if the loss of ranking loss branch should get more weight?
The second one is how much iters do you train, what is your strategy to tune the lr? (I use 4000 iters and Adam, and the lr divide by 10 on iter 1000, 2000, 3000)
The third question is do you use bias on every fc layer?
The last one is do you fix the bn layer in the backbone?
Hi, I noticed that in your paper on Table 5. The original embedding is 1536, but you reduce this dimension to 768/512 afterwards. I want to know that which method do you use in reducing dimension. Is this PCA or just simple reshape?
Looking forward to your reply. Thx!
Hello thanks for releasing this code. When you will release training code as well? Thanks
Hello guys!
I was wondering when you trained the model on the InShop dataset, what categories you used for calculating the classification loss?
Were they top-level categories like Blouses_Shirts, Cardigans etc. or were they more granular like id_00000001, id_00000271 etc.
I presume while the former can cohesively get broad categories together in the latent space, the latter can help guide the training to bring the same/similar clothes' embeddings together (as in red polo and blue polo together and so on). But I want to know how was the experiment conducted?
Can someone please help me with this issue?
Thanks & Regards.
Hi, could you tell me on which dataset your models are trained? Have you evaluated your models on other benchmarks like UKBench, Holidays, Oxford-5k, Paris-6k, .... If not, do you intent to test your models in order to compare your models with the SOTA models on these datasets.
Thank you in advance for your reply
Thank you for your great job of this paper!
1.Why can RESNET-50 get the best effect in CNN backbone network?
2.In the backbone network without down sampling, the input is 224x224x3, and the output is 14x14x1536, right?
Thank you for your great job of this repo and paper!
I notice that the result of ResNet-50 with non-conventional usage has the best performance. I want to know how to implement this ' non-conventional usage'.
Does it mean 'discarding the down-sampling operation between stage3 and stage4' in section 3.1 of the paper?
Thanks a lots.
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.