Panasonic vs Canon Autofocus

[MeanRevert]

Alrighty.. I'll say it only because I know someone else will...

Come on Panasonic, you have the words 'De-Focus' in the name!

 

 

[DBounce]

Poor AF is really Panasonic's main roadblock to greatness. Virtually everyone else has at least somewhat usable AF at this point. Panasonic has no excuses left. 

[Mako Sports]

1 hour ago, DBounce said:

Poor AF is really Panasonic's main roadblock to greatness. Virtually everyone else has at least somewhat usable AF at this point. Panasonic has no excuses left. 

I was pretty close to switching from Sony to Pany when  the GH5 1st dropped, but then I saw that AF performance lel

[webrunner5]

It is pretty good on the GH5 now. Son'y claim to fame is Face and Eye tracking. On average stuff it is not stand outish. Now on the A9, well sure that is great all around.

[Mako Sports]

This popped up on my feed, don't think it needs its own thread but C200 vs A6400 AF test

 

The test was v close and I was most surprised at how similar the colors were.

[webrunner5]

Both company's have come a Long ways. Pretty amazing how the price difference is between the two of them,  with close to the same results. The new Venice CS is pretty damn good now, and both are doing great with eye focus. Magic times we live in to have that kind of choice for so little money with the A6400.

[kye]

On 6/12/2019 at 11:07 AM, DBounce said:

Poor AF is really Panasonic's main roadblock to greatness. Virtually everyone else has at least somewhat usable AF at this point. Panasonic has no excuses left. 

I agree.

I think their theory that AI and image processing will eventually be the winning tech is probably correct, but we're likely a good amount of time away from that, enough certainly to implement phase detect and develop the DFD technology in parallel.  The two technologies kind of compliment each other in a way, with AI learning what should be the subject of the video and which direction the focus should be in, and PD confirming which direction to go for each area of the frame.  

Still, who knows what this stuff costs to implement, so maybe the numbers actually do stack up, they'll be laughing as they pull ahed of the other manufs when the DFD works well, and they'll have saved some money along the way, but who knows

[thebrothersthre3]

That rolling shutter jello tho

[BTM_Pix]

On 6/22/2019 at 5:35 PM, kye said:

I agree.

I think their theory that AI and image processing will eventually be the winning tech is probably correct, but we're likely a good amount of time away from that, enough certainly to implement phase detect and develop the DFD technology in parallel.  The two technologies kind of compliment each other in a way, with AI learning what should be the subject of the video and which direction the focus should be in, and PD confirming which direction to go for each area of the frame.  

Still, who knows what this stuff costs to implement, so maybe the numbers actually do stack up, they'll be laughing as they pull ahed of the other manufs when the DFD works well, and they'll have saved some money along the way, but who knows

It will be here much sooner than we expect.

This is a video of an Nvidia Jetson Nano SBC doing various real time image processing including tracking, mapping and even stabilisation.

The Jetson Nano is $99....

 

[kye]

40 minutes ago, BTM_Pix said:

It will be here much sooner than we expect.

This is a video of an Nvidia Jetson Nano SBC doing various real time image processing including tracking, mapping and even stabilisation.

The Jetson Nano is $99...

It will likely be here sooner than we expect, but I don't expect it for a decent amount of time, so even if it's a lot sooner it will still be a while

Those demos were sort-of interesting, but some were considered standard computer science graphics algorithms and were being given as assignments back in the early 90's.  I'd certainly hope that the processing power would be available to do such things, being that Moores law has had almost three decades to work on it!

We do have face-detection built in, which is a large part of it, but I recall that the harder part was getting computers to do object recognition in the less-than-perfect conditions.  I remember watching a 'latest tech' type show on TV many years ago and the guy was talking about computer vision, and how the computer could recognise a spoon pretty reliably, as well as many other kitchen cutlery they'd programmed into it, but when they pointed it at the drawer with all the random serving spoons and tin-openers that every house has, it was pretty much useless in partial matching things.  I'd suggest that there's a similar problem with recognising things in a 3D environment when they're not in focus, especially if you only have one lens to try and extrapolate depth from.  It's worth noting that the depth extraction algorithm was from a stereo pair of images, which probably had a very deep depth of field.  Humans, on the other hand are pretty good at depth perception using only one eye, using things like size, relative motion, parallax, contrast, and other factors to build a 3D model from a series of 2D images.

Maybe the AI of the future will have a few cameras on the front of the camera for analysis, or even like phones do where one is a low-res depth sensor.

 

[BTM_Pix]

11 hours ago, kye said:

It will likely be here sooner than we expect, but I don't expect it for a decent amount of time, so even if it's a lot sooner it will still be a while

The real time depth detection, tracking and object recognition demos I've seen with this of live video (HDMI>CSI converters) can form the basis of an AF system that I think will be integrated within a commercial product within 18 months.

Obviously It won't be this specific SBC as its just a consumer development/tinkering board (although it does itself have the potential to be the basis for an outboard system) but the principle will be the same.

How is that for a bold prediction