This Artificial Intelligence Powered Camera Describes What It Sees, With Its Own Voice (DIY project)

Artificial Intelligence

Machine learning algorithms are conquering the world. It’s not just now that they entered the photography domain.

This artificial intelligence application is a bit different from what comes in recent photo editing software. A Raspberry Pi is involved, along with a TensorFlow Lite object recognition software and Adafruit’s BrainCraft HAT machine learning system All together makes a very cool DIY project, described by Adafruit in an article.

[via DIY Photography]

Canon EOS 5D Mark III doing 4K video with Magic Lantern wasn’t an April Joke

I am sure most of you thought it was a clever April joke when a1ex of Magic Lantern team told the world on Saturday that they have made 4K video working on the EOS 5D Mark III.

Well, it turned out to be a very clever “April joke”, i.e. using the April joke hype to push a ground-breaking advancement of the Magic  Lantern hackers. Because that’s what it is: a ground-breaking advancement. Magic Lantern delivered the proof of concept that it is possible to get 4K video on an EOS 5D Mark III.

However, as cool as it is (it is), it’s a proof of concept, not even a working alpha version. There are many issues to solve before you can get working 4K on the EOS 5D Mark III. The main issue appears to be the fact that the EOS 5D Mark III can write to the memory card with max. 100MB/s, which will not allow for 4K@24fps (at least not more than a few seconds). The other big issue is surely overheating.

And there is also another exciting news. The EOS 5D Mark III may never do professional 4K video, but it may soon do 3K video with 24fps and 14 bit – this should produce a data rate of approx. 90MB/s, doable for the EOS 5D Mark III’s Output pipeline. Well, if these aren’t exciting news.

These are the video modes that can be activated with the latest ML hack:

  • 1920×960 @ 50p (both 1:1 crop and full-frame – 3×3 pixel binning)
  • 1920×800 @ 60p (same as above)
  • 1920×1080 @ 45p and 48p (3×3 binning)
  • 1920×1920 @ 24p (1:1 square crop)
  • 3072×1920 @ 24p (1:1 crop)
  • 3840×1600 @ 24p (1:1 crop)
  • 4096×2560 @ 12.5p (1:1 crop)
  • 4096×1440 @ 25p (1:1 crop)
  • Full-resolution LiveView: 5796×3870 at 7.4 fps (128ms rolling shutter).

A1ex writes:

This is only a very rough proof of concept. It has not been battle-tested and has many quirks. Some of them may be easy to fix, others not so. In particular:

* It feels quite buggy. I’m still hunting the issues one by one, but it’s hard, as Canon’s LiveView implementation is very complex, and our understanding on how it works is still very limited.

* Write speeds are high. For example, 10-bit 4096×2500 at 15 fps requires 180 MB/s. 1080p45 should be a little more manageable at 111 MB/s.

* Canon preview is broken in most modes; you need to use the grayscale preview in the raw recording module.

* High-resolution modes (in particular, full-res LiveView) may cause trouble with memory management. This is very tricky to solve, as we only get 3 full-resolution buffers in LiveView, with restrictions on the order in which they must be freed, and lots of other quirks.

* Since these settings were pushed to limit, the risk of corrupted frames is high. If it happens, decrease the vertical resolution a bit.

* When refreshing LiveView settings, the camera might lock-up (no idea why). Pressing MENU twice appears to fix it.

What do you think? Isn’t this exciting, and incredibly cool? To learn more about the latest Magic Lantern hack, head over to the ML forum or visit the website.