Exercise 4: Six at once (PR135)¶

You will tell mcdonald how many objects you are looking for, and see what it does with the number: here it takes a flock apart, and gives each bird a report of its own.

Time: about 35 to 45 minutes, nearly all of it the computer working.
Space: a 23 MB download, and about 50 MB for the frames you open.
You need: mcdonald 0.3.8 or newer, installed and set up (Install).

The video¶

PR135 is one minute of infrared video from a US military camera, filmed in 2025. The report that came with it describes "six small spherical objects grouped together" that "seem[ingly]" were "agile" and "frequently changing direction", and the person who reported them put their speed at 480 miles an hour. The record says the camera turns to follow them from 0:11 to 0:47. You can read the whole record on its DVIDS page. You will look at five seconds of it, from 0:41.34 to 0:46.31.

The report says six, so this time you tell mcdonald that you are looking for six objects.

In the window¶

Start the window: type mcdonald-gui in a terminal, or use the mcdonald icon if mcdonald setup added one.
On the first page, press Open by catalog name…, type PR135, and press OK.
mcdonald asks whether to download PR135 from DVIDS. Press Yes.
The page Select the video segment opens. Click in the box beside Set start, type 1240 and press Enter. Then click in the box beside Set end, type 1389 and press Enter. The line under the boxes should say 0:41.34 to 0:46.31. Press Play segment to watch it: two small clusters of bright spots, below and to the left of the middle, drift to the left together.
Now the number. Click in the box beside How many objects are you looking for?, where it says I don't know, type 6 and press Enter. Then press Open this segment.
mcdonald saves those frames as pictures, with a bar on the right that shows how far it has got.
In the Measure the object box on the right, press Begin. Three lights under the button show the steps: Find, Follow and Measure. Nothing is asked on the way. This takes about 30 to 40 minutes: mcdonald measures each thing it found, and then each bird on its own.
When it ends, the line under the button holds your number against what was followed, and the right column has a report card for each object. Scroll down to see them all.

On the command line¶

These are the same steps, for a terminal or an AI agent. First, find what moves in those frames, and say how many objects you are looking for:

mcdonald look PR135 --n0 1240 --n1 1389 --propose --objects 6

This downloads PR135 the first time, saves the frames, and looks for things that move against the background. It takes about 3 minutes. It lists 3 things, best first, and draws them in the picture pr135/pr135_look_proposals_1240_1389.png. Thing 1 is "strong", thing 2 "fair" and thing 3 "weak". Things 1 and 2 are the two clusters, seen on 87 and 82 frames; thing 3 was seen on only 3. Each line says the thing "holds about 3 points" or "about 2 points, not one": it may be a group. Then it says that you are looking for 6 objects and 3 things were found, and that a group of points can be split into its members.

Under the list, it prints to take the 3 likeliest: and three commands, one for each thing. Copy each one, replace the ... at the end with your own reason, and add --out pr135/object-1 to the first, --out pr135/object-2 to the second and --out pr135/object-3 to the third, so that each thing has a folder of its own. Here they are, shortened, with that added:

mcdonald mark PR135 --no-window --link --set object@1281=420.0,635.0 … --set object@1387=257.0,640.0 --why "proposal 1 of 3 from look --propose (…); I looked at its strip and it is the object because ..." --out pr135/object-1
mcdonald mark PR135 --no-window --link --set object@1256=381.0,578.0 … --set object@1351=121.0,596.0 --why "proposal 2 of 3 from look --propose (…); I looked at its strip and it is the object because ..." --out pr135/object-2
mcdonald mark PR135 --no-window --link --set object@1242=428.0,518.0 --set object@1246=433.0,549.0 --why "proposal 3 of 3 from look --propose (…); I looked at its strip and it is the object because ..." --out pr135/object-3

Run each, in turn. They should say "151 of 151 frames linked, 1251–1401", "156 of 156 frames linked, 1226–1381" and "65 of 65 frames linked, 1212–1276". Marks set this way are recorded as an agent's choice, with your reason, and the reports say so. That is expected.

Then measure them all, over the same frames, with the same number. This takes about 30 minutes:

mcdonald run PR135 --each pr135 --objects 6 --n0 1240 --n1 1389 --skip integrity

--skip integrity leaves out the slowest check, as the window does. The list of the objects and what each report says is pr135/pr135_objects.md.

To have an AI agent do it, you can give it this prompt:

Please use the mcdonald toolkit to analyze the PR135 video released under PURSUE, following https://github.com/haqqmisra/mcdonald/blob/main/docs/exercises/04-pr135.md. Run mcdonald readme to get started.

What you should get¶

Both ways give the same reports. The line under the button, and the first line of the list, say:

You looked for 6 objects: 6 were followed, 6 of them the members of 2 groups (objects 1, 2); object 3 followed the same thing as object 8.

Objects 1 and 2 are the two groups, each measured as a flock. Their reports start:

Beats like a bird. A flock of birds is the leading explanation: its 3 members each beat at 6.9–7.8 Hz, the rate of a wingbeat, each its own and out of step with the others (a rhythm of the video would beat them as one; tumbling bodies or blinking lights would beat too; 3 more were seen too briefly to tell); the object moved about 106 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A flock of birds is the leading explanation: its 3 members each beat at 7.5–7.9 Hz, the rate of a wingbeat, each its own and out of step with the others (a rhythm of the video would beat them as one; tumbling bodies or blinking lights would beat too; 1 more was seen too briefly to tell); the object moved about 106 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Objects 4 to 9 are the six birds, each followed on its own, three from each group. Their reports start:

Beats like a bird. A bird is the leading explanation: its brightness beats at 7.8 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 75 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A bird is the leading explanation: its brightness beats at 7.5 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 115 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A bird is the leading explanation: its brightness beats at 6.9 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 82 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A bird is the leading explanation: its brightness beats at 7.9 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 99 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A bird is the leading explanation: its brightness beats at 7.8 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 86 pixels a second against the background, and its real speed and size cannot be found from this video alone.

Beats like a bird. A bird is the leading explanation: its brightness beats at 7.5 Hz, the rate of a wingbeat, a rhythm of its own and not the video's (a tumbling body or a blinking light would beat too, and nothing here tells them apart); the object moved about 85 pixels a second against the background, and its real speed and size cannot be found from this video alone.

In the window there are 8 report cards, and the video shows each object's track in its own colour. Object 3 has no card: it is the same bird as object 8, as the next part says.

What it means¶

The number changes what mcdonald does. Find saw three things: two that each hold a few points, and a third seen on only 3 frames. Left at I don't know, the one button follows the two groups and measures each as one flock, in about 11 minutes, and you get two report cards. Told 6, it takes every thing Find listed, weak ones too. It found fewer things than 6, and two of them are groups, so each point of a group that was seen long enough for a beat becomes an object of its own. That is how you get a report for each bird.
Six birds. Each of objects 4 to 9 beats its brightness at its own rate, between 6.9 and 7.9 times a second, out of step with the others. That is the rate of a small bird's wingbeat. An earlier analysis of this video found the same: six birds, each beating its wings about 7 to 8 times a second. The flock's reports say why it cannot be the video's own rhythm: that would beat every bird at once.
Counted once. Object 3 is the third thing Find saw, on 3 frames. Followed from there, it is one of the second group's birds, seen before that bird was followed on its own as object 8: on every frame both were followed on, the two are on the same pixels. So it is the same bird, followed twice, and mcdonald counts it once. Its report is still in its folder, and its part of the list says why it is not counted.
The number is a hint, not an order. If mcdonald finds fewer things than you said, it tells you how many were not found. If it finds more, it tells you that too. It never makes up a thing to reach your number.
Not lanterns, as far as this goes. Press Full report on object 2. A sky lantern's flame flickers at rates like these too, but lanterns let go together drift on one wind and keep their places. This group does not keep its shape: it moves around the way a flock does.
The speed is only on the screen. The camera follows the birds, and nothing in the video says how far away they were. The report says what it would need.
Check the tracks by eye. Each report card says Track not yet checked by eye. Press Full report on each, find its track sheet, and if the ring is on the bird in every frame, press I looked. On the command line, open each pr135/object-N/pr135_all_frames.jpg, then run mcdonald report pr135/object-4/pr135_case.json --i-looked for object 4, and the same for the others.

Tell us how it went¶

Please send us your result, whether or not it matched. Tell us what the line under the button said, how many report cards you got, how long it took, and anything that was hard to follow.