Thursday, 29 May 2014

GH4 focus performance during video recording

Recent Micro Four Thirds cameras have very good autofocus performance for still images. Mostly, the performance is among the best in this class, certainly better than DSLR cameras in live view mode. However, there is one area where mirrorless cameras don't perform well at the moment, and that is continuous autofocus: Both during video recording, and for photographing moving objects, e.g., for photographing sports and birds.

Some camera manufacturers have been trying to solve this by adding phase difference sensors (PDAF) on the imaging chip, like the Nikon 1 and Sony E-mount cameras. However, the real world benefit of that solution is still somewhat undecided. The Nikon 1 cameras appear best in this respect so far.

Panasonic have said in interviews that the on-sensor PDAF solution is not going to be used for their Micro Four Thirds cameras, at least not anytime soon. Rather, Panasonic expects to achieve better continuous autofocus performance by using faster image readout from the chip, better image processing algorithms, and more processing power. Have they achieved this with the most recent Lumix GH4?

With Lumix X 12-35mm f/2.8 @ 35mm f/2.8

To compare the autofocus performance during video recording with the GH4, I used a Lego Technic contraption to move a cardboard box back and forth at a steady pace. I then set up both cameras, in turn, with the Lumix X 12-35mm f/2.8 lens at 35mm f/2.8 at close range, and recorded video at 1920x1080, progressive, 50fps. With the Lumix GH4, I also recorded 4K video footage at 25fps.

I used the Manfrotto 190XPROB tripod, which has a centre column that can easily be set horizontally, very useful for this use. As the box moves, the shortest focus distance is 0.36m (just inside the lens close focus distance), and the longest focus distance is 0.5m.

Comparing the resulting footage, it is easy to see which camera better finds the focus during the movement. Here is the video footage, for comparison:

I used the exposure settings: ISO 400, f/2.8, 1/100s. The 4K footage was scaled down to 1080 rows.

It's easy to see that the GH4 achieves correct focus more often than the GH3, in fact, about twice as often, according to my crude frame counting.

At 4K, the focus is quite poor. This is consistent with my previous experience, see for the video example in this article about Kdenlive, a free 4K video editing software. In the example video, you can see that zooming while recording 4K video loses the focus, much more so than when recording 1080p footage.

I'm guessing there are two reasons for the worse focus performance during 4K video: One is that the camera is only recording 25 frames per second. The lower number of frames means that the camera cannot adjust the focus as fast. It can only base the focus correctness on each frame recorded, and the lower the frame rate, the slower the focus can move.

Second, the GH4 is probably using much more processing power to capture the 4K video stream, leaving less to the autofocus processing.

Both the Lumix GH4 and the GH3 support the 240fps contrast detection autofocus (CDAF) sensor readout with some select lenses. So far, this is only possible when using the newest f/2.8 zoom lenses, the Lumix X 12-35mm f/2.8 and the Lumix X 35mm-100 f/2.8.

But can the camera use the 240fps feature during video recording? Probably not, since the sensor is busy reading the image at 50fps for the video stream anyway.

I used a shutter speed of 1/100s. This means that there is an "unused" 1/100s for each frame, which the camera may be able to use for AF, depending on the sensor and processor design. Whether Panasonic is able to exploit these unused exposure segments is probably unknown and a business secret.

With Lumix X 35-100mm f/2.8 @ 100mm f/2.8

Here is the same test with the Lumix X 35mm-100 f/2.8 at 100mm f/2.8. It was done later in the evening, so I had to up the ISO to 1600, otherwise I used the same exposure settings. Also, the focus distance was increased to around 1m.

We see pretty much the same thing here: The Lumix GH4 does better, but perhaps not vastly better. Also, when doing 4K video, the autofocus is pretty much useless in this test.


The Lumix GH4 further improves upon the GH3 in terms of autofocus speed in 1080p mode. However, the autofocus performance during video recording is quite poor when recording 4K video. So if you absolutely need to rely on autofocus, use 1080p video at a high framerate, rather than 4K video. In 4K mode, it may pay to prefocus, and then turn off autofocus while recording videos.

1 comment:

  1. It may be a good thing to slow this lego movements down.

    It's hard to tell if focussing is a result of acquiring or hunting, or just because the object happens to be in focus at some moment.

    If each situation leaves room (time) for actually acquiring focus, the comparison may be more valid.

    Perhaps you should rotate in a flat space (circle horizontally), instead of a vertical area.