[torotonnato]

I don't think they are using Photoshop, but that would be hilarious :D It's just the camera sensor's color space (Adobe RGB). Anyway, gamma correction should be problem, unless they shoot raw images and in this case the color space gets out of the equation.

I was curious, so I found another paper by the original authors and the color correction is more carefully considered, since they seem to take into account the sensor response too [0].

[0] - https://arxiv.org/pdf/2106.07403.pdf

[jiggawatts]

This paper also doesn't mention "gamma" and uses the word "linear" only in other contexts.

Even if they're using RAW photos, the response curve is still non-linear because the individual pixels "saturate" as they get closer to the maximum exposure. This shifts colours, because a bright colourful source will saturate the pixels of the matching colour first, and then the other colours a bit later. A bright yellow meteor trail will saturate red and green, and then blue.

Their entire method and conclusion all hinges on analysing the relative intensities of RGB colours of photos of very bright meteor trails.

These guys are so unscientific it's almost a parody of science. It reads like a bunch of high school students doing "science" with their dad's camera, and then a kind professor submits their homework to arXiv to make them feel like they're Just Like The Big Boys.

Having said that, there is some amazing real scientific research being done with CotS camera equipment!

Examples:

https://astronomytechnologytoday.com/2018/06/28/miniwasp-par...

https://petapixel.com/2022/07/25/telescope-made-from-multipl...

https://www.osti.gov/servlets/purl/1561833

That last one is a beautiful example of how to do science right, with detailed calibration data and characterisation of every aberration in the system.

[torotonnato]

>These guys are so unscientific...

Naaaaa, this is not my field, so I assume I'm missing some crucial pieces of information (which the paper apparently lack too, I agree on that).

>A bright yellow meteor trail will saturate red and green, and then blue.

Here I assume that they can do their job, but maybe you are right. Since they are doing their observations at two "meteor stations", maybe they know what they are talking about though.

>Their entire method and conclusion all hinges on analisying the relative intensities of RGB colours of photos of very bright meteor trails.

They use a simple mathematical model, with even other obvious limitations, like considering the atmosphere homogenous, so?

Here I'm citing the calibration part (which is a bit disappointing, true :) at the end of page 4:

...The color chart in Fig. 10 allows us to evaluate the color characteristics of the Moon and check the calibration of our cameras. The Moon has a color relative to the sky background: B - G = -2.5 log (1.7 / 2.7) = 0.5. We take into account the color correction in the Jhonson B - V system according to [x] due to Rayleigh scattering equal to 0.14 magnitude. Let’s get the estimate B - V of the Moon: B - V = 0.50 + 0.60 - 0.14 = 0.96. The actual color of the Moon is B - V = 0.91 according to [1] and differs from our estimate by 0.05 magnitudes within the photometric error. In Figure 9 we can see a local feature (water tower). The color diagram of the tower in Fig. 12 gives a distance estimate of 0 ± 1 km. The actual distance is about 300 meters. Thus, colorimetric measurements confirm our estimates...

EDIT: page 4 of the UAP paper

[jiggawatts]

They're not accounting for exposure levels anywhere that I can see, but to be honest I just skimmed it quickly looking for the expected formulas.

It's easy to write garbage in the style of a scientific paper. Use LaTeX, the right fonts, layout, and tone and most people will immediately stop questioning the content. Pepper it with formulas and it starts to look like wisdom delivered by serious men in white lab coats.

I studied physics with a particular focus on optics because I wanted (and achieved) a career in computer graphics. I wanted a good solid background on light transport and physically accurate rendering techniques.

This paper covers about 5% of what you would expect to see in a serious publication, if that. The calibration "technique" they use is hilariously bad. The assumptions are invalid. Obvious instrument limitations aren't even mentioned let alone corrected for.

Compare to the last link in my previous post. The difference is night and day.

Similarly, for a vaguely related recent example, take a look at how the JWST telescope calibration is done: https://jwst-pipeline.readthedocs.io/en/latest/jwst/linearit...

Take a look at the menu on the left! There's section after section after section that covers every aspect of this instrument! Fundamentally, it's "just" a fancy camera with a CCD/CMOS sensor, optics, and similar limits. It's the same problem space, so it's a good example of how this can be done properly.

Admittedly, JWST has a huge budget, but even amateurs do similar things when performing "image stacking", and that's just for making pretty pictures, not for scientific publication: http://deepskystacker.free.fr/english/theory.htm#Calibration

If I were given a budget of just a few thousand dollars and a year to write a PhD or something, I would:

- Get a few identical cameras and lenses, ideally prime lenses. Budget permitting, the best kind are the B&W digital ones, like this one: https://leica-camera.com/en-AU/photography/cameras/m/m10-mon...

- Mount three or more of them on something sturdy facing the same patch of sky, at least a hundred meters apart in a large triangle or other similar shape. Ideally, several at each spot aligned in parallel but using different, well-characterised filters. Or a single camera with rapidly swappable filters, which would be within reach of a student's budget using 3D printing.

- At each location have a GPS receiver as a timing source. Handheld receivers suffice, and many Arduino-type boards have modules for this. Also have a "weather station" measuring wind speed, humidity, and temperature. Also have a thermal probe stuck to the sensor of the camera, or very near it.

- Perform detailed lab calibration of each camera and lens pair. Repeat for each filter if using filters. Use a proper instrument like a spectrophotometer, which every physics lab will most likely have lying around somewhere. This need to be performed at the same focus ("infinity") and across a range of intensities. Ideally at different temperatures also.

- Calibrate in the field. E.g.: align the photos using stars or perform similar cross-checks between each camera. Use GPS data relayed from airliners to verify the altitude computation from stereopsis. Use aircraft that you know are "white" to cross-check atmospheric conditions. Fly drones up a few kilometres with blinking LED lasers on them of various wavelengths. Etc...

- Use manual focus, synchronized camera settings, and take photos every second or so using GPS timing for accurate sync. Ensure to stay well within the "middle" of the dynamic range of the sensors.

- Collect the information for analysis, ideally over weeks or months. Multiple cameras with different perspectives allows massively more accurate estimates of a range of parameters, especially height, speed, size, etc... It simultaneously helps eliminate spurious sensor problems, bugs, birds, raindrops, and other confounding factors.

This could be done with a few thousand dollars and the data could keep many students busy publishing good papers for years. You could write papers on cloud formation, bird migration, raptor hunting statistics, accuracy of aircraft transponders, lightning frequency, meteors, satellites, and even cover military topics such as spotting drones! Computer Science students could get involved with AI analysis of the data, efficient storage, real-time analysis and tracking with longer telephoto lenses, etc...

This is what real science looks like, even if done on a budget. There are teams out there basically doing this or some variant right now!

[torotonnato]

Absolutely 100% agree with you on how the calibration could be done and that the methods seems sloppy. There are lots of implicit assumptions (but maybe the paper is less lacking for those people doing similar jobs).

The way I interpreted the pre-print:

- They optimized the capture for very fast moving objects (and large FOVs? Maybe using a spectrophotometer is not as straightforward as it seems)

- They are making estimates more than precise measurements, using error bars that are "good enough" for them, for the specific domain and taking into consideration the constraints given by the assumptions and the equipment. Not that I see those errors bars, though :D

- Now that I'm re-reading: ...Frames were recorded using the ser format with 14 and 16 bits... (ser is an astronomical file format)

- Paper's exposition could be better

- They are not building the JWST. They are doing something like lambertian reflectance when you are asking for global illumination. Let's wait for (the negative results of) NASA's study. I share your skepticism btw, not a believer.

Just saying that they are not using Photoshop. The paper is lacking in details though, that's uncontroversial.

Been a pleasure talking with you, have a nice day!

[retcore]

>It's just the camera sensor's color space (Adobe RGB)

Sensors don't have a color space. It's a transformation applied in the camera software. Unfortunately no commercially available camera either exposes using the sensor data output itself, leading to often considerable non linearity between the measurable output values and the actual exposure. This is applied to the raw file for exposure and post processing raw conversion. You need to analyse the non de-bayered raw files directly to assess what's happening in the captured data.