search

BelalC / sign2text

Real-time AI-powered translation of American Sign Language fingerspelling (sign to text)
155 stars 48 forks source link

Running command #4

Closed joohaeng closed 6 years ago

joohaeng commented 6 years ago

@BelalC, When running the demo, I specify the name of the pre-trained net using "-m" option. But, I could not make it work using "-weights" option. Is it related to the files in "weights" folder? Could you give some working example?

BelalC commented 6 years ago

@joohaeng - yes the '-w' argument is for specifying what weights you want to load into the model. By default, the model will load the pre-trained weights I included in the 'weights' folder. You only need to specify the weights argument if you want to load your own pre-trained weights. Example using weights included in repo:

python live_demo.py -m vgg16

joohaeng commented 6 years ago

@BelalC Thanks! By the way, which pre-trained net shows the best performance?

BelalC commented 6 years ago

VGG16 was the most accurate the last time I tested it out, but it's fairly compute heavy. MobileNet was the fastest but least accurate, although it could definitely be improved with some more tweaking + data. Let me know how you go!