There are a bunch of constraints with updating neural networks. EG- Probably training the network with animal images is doable and they have the sensor/processor margins to do it. I’m less sure what’s possible through OVF capability expansion, for example.
Anyway, we don’t know how Canon’s various IC workloads are parsed, what they’re doing to improve the efficiency of the stack, where the system bottlenecks are etc, hence my caveats.
There are a bunch of constraints with updating neural networks. EG- Probably training the network with animal images is doable and they have the sensor/processor margins to do it. I’m less sure what’s possible through OVF capability expansion. Anyway, we don’t know how Canon’s various IC workloads are parsed, what they’re doing to improve the efficiency of the stack, where the system bottlenecks are etc, hence my caveats.
Mar 02, 2020 at 09:12 AM
Previous versions of Dlang's message #15159848 « Official 1dx Mark III picture thread »
