search

liyunqianggyn / Deep-Unsupervised-Image-Hashing

Implementation of accepted AAAI 2021 paper: Deep Unsupervised Image Hashing by Maximizing Bit Entropy
80 stars 9 forks source link

The result is different form paper #5

Closed Lukangkang123 closed 2 years ago

Lukangkang123 commented 3 years ago

I tested on the cifar10 dataset, but found that the map was only 0.22 8481089B53758DE622056967A6C2B83B

liyunqianggyn commented 3 years ago

Thanks for your interests! Seems the training loss does not converge well at the end. I will run the code to check it and I'll leave another comment here when I have. Please stay tuned! :)

liyunqianggyn commented 3 years ago

Hello I run the code with same setting as in released code, "train_dataset 50000 test_dataset 10000 database_dataset 50000 Lr = 0.0001 bit = 64 gamma = 6" The mAP can reach 59.6% (in paper it is 59.5%) as far. Here is the training log results: image image

May I ask which implementation code you are using? Yesterday we found a small mistake in another implementation "DeepHash-pytorch", if you are using it, please refer to the issue1 and issue2 I've also email this small problem, and the problem has been corrected now.

Hope it can help you! Please let me know if you have other questions, this is helping us to rethink more of our paper :)

Lukangkang123 commented 3 years ago

Hello I run the code with same setting as in released code, "train_dataset 50000 test_dataset 10000 database_dataset 50000 Lr = 0.0001 bit = 64 gamma = 6" The mAP can reach 59.6% (in paper it is 59.5%) as far. Here is the training log results: image image

May I ask which implementation code you are using? Yesterday we found a small mistake in another implementation "DeepHash-pytorch", if you are using it, please refer to the issue1 and issue2 I've also email this small problem, and the problem has been corrected now.

Hope it can help you! Please let me know if you have other questions, this is helping us to rethink more of our paper :)

I downloaded and ran the code again, then got the result of 0.5 +. Thanks you for your reply.