FasterNet

FasterNet copied to clipboard

some question

3

Can I use 416*416 picture as input？ what about 448*448 in detection without pre-training model?Could fasternet as YOLOV3 backbone?

jackiezhang11

why not use activation functions after downsampling convolutions？

1

Great work! But why not use activation functions after downsampling convolutions?

1920230345

The order of the blocks in the Stage

1

hi，in your paper, for a stage，the PConv is before PWconv+BN+Act+PWconv, but in the code , the order is reversed?  the paper  the code

wsy-yjys

A large pre-training weight

2

nice work，Can you train a large network of 128,256,512,1048?

zhu011

About PWConv

1

hello,just as the paper showing , A PConv followed by a PWConv,but I can't find the PWConv in your code.can you help me? thank you!

xyh001007

Can it be used on mmdetection？

1

okt7760

The difference between 'slicing' mode and 'split cat' mode?

1

Hello, thank you for your simple but powerful work. There is a question about the Partial_conv3. I test the inference speed with 'slicing' mode and 'split cat' mode and find...

cswry

why the model size all so big?

5

compared with mbv2 and shufflenetv2 fasternet has a bigger size, this caused limitations deploy on edge devices.

lucasjinreal

Some problem 2

1

Are there any examples of PConv migrating to other detection frameworks? For example, YoloV5, YOLOV8

wemindful

some question about reparameterization in pconv

1

 When using this function to combine pconv and bn，Because pconv uses partial dimensions for the volume machine。This causes the dimensions of kernel and t to be inconsistent。What should I...

870231652