search

naver / r2d2

Other
461 stars 86 forks source link

Confusion about the different architectures model size #54

Closed ShowStopperTheSecond closed 3 months ago

ShowStopperTheSecond commented 3 months ago

Hi there

In the Readme file there are details about some of the pretrained models. There is a column about the network size (# of parameters).

model name model size
(#weights)		 number of
keypoints		 MMA@3 on
HPatches
r2d2_WAF_N16.pt 0.5M 5K 0.686
r2d2_WASF_N16.pt 0.5M 5K 0.721
r2d2_WASF_N8_big.pt 1.0M 10K 0.692
faster2d2_WASF_N8_big.pt 1.0M 5K 0.650
<!-- r2d2_WASF_N8_big.pt 1.0M 5K 0.704 -->

When I check the code the number of parameters for the Fast-R2D2 is about 0.5M not 1.0M. Please clarify this. thanks

ShowStopperTheSecond commented 3 months ago

N is a parameter of the model architecture too