Hi, I noticed that in Table 5, the accuracy of training from scratch is only 70.9% for

Hi <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="

Hi <a class="user-mention notranslate" data-hovercard-type="user" data-ho

Hi <a class="user-mention notranslate" data-hovercard-type="

CIFAR100 Results about s2-bnn HOT 3 CLOSED

NicoNico6 commented on May 27, 2024 1

CIFAR100 Results

from s2-bnn.

Comments (3)

szq0214 commented on May 27, 2024

Hi @NicoNico6, thanks very much for your interest in our work! It seems the vanilla MobileNet itself performs not well on CIFAR-100.
https://github.com/NoUnique/MobileNet-CIFAR100.pytorch
https://github.com/weiaicunzai/pytorch-cifar100
This may be caused by its structural property but I think it definitely can be further studied in depth.

from s2-bnn.

Xingrun-Xing commented on May 27, 2024

Hi @NicoNico6, thanks very much for your interest in our work! It seems the vanilla MobileNet itself performs not well on CIFAR-100. https://github.com/NoUnique/MobileNet-CIFAR100.pytorch https://github.com/weiaicunzai/pytorch-cifar100 This may be caused by its structural property but I think it definitely can be further studied in depth.

你好，cifar10和100你们用多大分辨率训的，另外二值网络from scratch是指reactnet one step的训练结果吗？多谢

from s2-bnn.

szq0214 commented on May 27, 2024

Hi @NicoNico6, thanks very much for your interest in our work! It seems the vanilla MobileNet itself performs not well on CIFAR-100. https://github.com/NoUnique/MobileNet-CIFAR100.pytorch https://github.com/weiaicunzai/pytorch-cifar100 This may be caused by its structural property but I think it definitely can be further studied in depth.

你好，cifar10和100你们用多大分辨率训的，另外二值网络from scratch是指reactnet one step的训练结果吗？多谢

Hi @Xingrun-Xing 分辨率是224x224的因为要match ImageNet上的模型 (finetuning和linear evaluation)，from scratch指one step training。

from s2-bnn.

Recommend Projects