The AI in the Sony cameras refers to a pre-trained neural network. The weights of the network are set and not adjustable by the user... the AI chip does the elaboration based on the inputs not the training . training requires GPU like horse power and is done offline during product development.
from late 1990's till 2010's Canon used to call their continuous servo "AI servo" AI stood for "artificial intelligence" of course those cameras had no ML, neural network or anything remotely close to that.... just rigid one size fit all algorithms... so by that token Sony is legit AI.... but it's not same level of AI as in a Tesla self driving tech for example. it could even be a simple regression model rather than convolutional neural network.
marketing aside, It's not trivial to make a camera that can "learn" from a user in the field, for example in terms of training AF, if the camera misses focus and ends up with a total blur, how is the user going to classify what it should have focused on after the fact using a total OOF frame? it's not that simple in this case. to re-optimize those weights one would need to feed in specific images of the novel subject with variable focus distances and other criteria...
Training AWB in the field might be possible by the user if they had access to calibrated targets but why would anyone want that ? We pay Sony to do it for us and if shooting RAW, the same training can be done in the RAW conversion software AWB profile where it is much more effective with the computer horsepower behind it.... and each individual can train the profile to their liking...
so not sure what a camera that "learns" in the field from the "photographer" even means... ? camera is a tool and not meant to become a photographer by itself...
the new camera seems to be a nice addition with 30 fps blackout free shooting.
The AI in the Sony cameras refers to a pre-trained neural network. The weights of the network are set and not adjustable by the user... the AI chip does the elaboration based on the inputs
from late 1990's till 2010's Canon used to call their continuous servo "AI servo" AI stood for "artificial intelligence" of course those cameras had no ML, neural network or anything remotely close to that.... just rigid one size fit all algorithms... so by that token Sony is legit AI.... but it's not same level of AI as in a Tesla self driving tech for example. it could even be a simple regression model rather than convolutional neural network.
marketing aside, It's not trivial to make a camera that can "learn" from a user in the field, for example in terms of training AF, if the camera misses focus and ends up with a total blur, how is the user going to classify what it should have focused on after the fact using a total OOF frame? it's not that simple in this case. to re-optimize those weights one would need to feed in specific images of the novel subject with variable focus distances and other criteria...
Training AWB in the field might be possible by the user if they had access to calibrated targets but why would anyone want that ? We pay Sony to do it for us and if shooting RAW, the same training can be done in the RAW conversion software AWB profile where it is much more effective with the computer horsepower behind it.... and each individual can train the profile to their liking...
so not sure what a camera that "learns" in the field from the "photographer" even means... ? camera is a tool and not meant to become a photographer by itself...
the new camera seems to be a nice addition with 30 fps blackout free shooting.
The AI in the Sony cameras refers to a pre-trained neural network. The weights of the network are set and not adjustable by the user... the AI chip does the elaboration based on the inputs
from late 1990's till 2010's Canon used to call their continuous servo "AI servo" AI stood for "artificial intelligence" of course those cameras had no ML, neural network or anything remotely close to that.... just rigid one size fit all algorithms... so by that token Sony is legit AI.... but it's not same level of AI as in a Tesla self driving tech for example. it could even be a simple regression model rather than convolutional neural network.
marketing aside, It's not trivial to make a camera that can "learn" from a user in the field, for example in terms of training AF, if the camera misses focus and ends up with a total blur, how is the user going to classify what it should have focused on after the fact using a total OOF frame? it's not that simple in this case. to re-optimize those weights one would need to feed in specific images of the novel subject with variable focus distances and other criteria...
Training AWB in the field might be possible by the user if they had access to calibrated targets but why would anyone want that ? We pay Sony to do it for us and if shooting RAW, the same training can be done in the RAW conversion software AWB profile where it is much more effective with the computer horsepower behind it.... and each individual can train the profile to their liking...
so not sure what a camera that "learns" in the filed from the "photographer" even means... ?
the new camera seems to be a nice addition with 30 fps blackout free shooting.
The AI in the Sony cameras refers to a pre-trained neural network. The weights of the network are set and not adjustable by the user... the AI chip does the elaboration based on the inputs
from late 1990's till 2010's Canon used to call their continuous servo "AI servo" AI stood for "artificial intelligence" of course those cameras had no ML, neural network or anything remotely close to that.... just rigid one size fit all algorithms... so by that token Sony is legit AI.... but it's not same level of AI as in a Tesla self driving tech for examples
It's not trivial to make a camera that can "learn" from a user in the field, for example in terms of training AF, if the camera misses focus and ends up with a total blur, how is the user going to classify what it should have focused on after the fact using a total OOF frame? it's not that simple in this case. to re-optimize those weights one would need to feed in specific images of a novel subject with variable focus distances and other criteria.
Training AWB in the field might be possible by the user if they had access to calibrated targets but why would anyone want that ? We pay Sony to do it for us and if shooting RAW, the same training can be done in the RAW conversion software AWB profile where it is much more effective with the computer horsepower behind it.... and each individual can train the profile to their liking...
so not sure what a camera that "learns" in the filed from the "photographer" even means... ?
the new camera seems to be a nice addition with 30 fps blackout free shooting.
Dec 03, 2025 at 03:02 AM
Previous versions of speedmaster20d's message #16940769 « A7V Announcement »
