Change backbone to mobilenetv3

MVIG-SJTU / AlphaPose

Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System

http://mvig.org/research/alphapose.html

Other

7.93k stars 1.97k forks source link

Change backbone to mobilenetv3 #1010

Open dekathomas opened 2 years ago

dekathomas commented 2 years ago

Hello @Fang-Haoshu @HaoyiZhu, I have tried to update the torchvision and using mobilenet to train alphapose, bu I got error like this

I change file alphpapose/models/fastpose.py at line 29 to self.preact = eval(f"tm.mobilenetv3_large()") and 34 x = eval(f"tm.mobilenetv3_large()")

Could you help me? Thank you in advance

dekathomas commented 2 years ago

Could you help me with this problem please? @Fang-Haoshu @HaoyiZhu @Jeff-sjtu

HaoyiZhu commented 2 years ago

I'm not familiar with mobilenet, but it seems the dim of input to pixel_shuffle is wrong. Could you print it's shape out?