Siberian Jay Feeder Ledger
videos/ holds four real field recordings from a long-term study of wild Siberian jays in
northern Sweden: feeder_1.mp4 … feeder_4.mp4. Each is a single continuous take from a
fixed camera pointed at a wooden feeding perch: an upright post with two horizontal poles,
each with a bundle of food wrapped in light-coloured cloth tied at the middle. Jays fly in,
perch, peck at the food bundles, and leave, often several at once. The files are 1920×1080
at 25 fps, 15 to 31 minutes long, with no audio. Fast motion shows interlacing artifacts.
Identifying birds
Each jay carries four leg rings, two on each leg, one above the other. A bird's code is four
letters in this order: top ring on its left leg, bottom ring on its left leg, top ring on
its right leg, bottom ring on its right leg. Left and right are the bird's own left and
right, not the image's (a bird facing the camera shows its left leg on the image's right).
| letter |
ring colour |
letter |
ring colour |
a |
aluminium (metal) |
o |
orange |
c |
chrome (grey) |
y |
yellow |
g |
green |
s |
sky (light blue) |
l |
lime (light green) |
b |
blue |
r |
red |
m |
mauve (purple) |
w |
white |
p |
pink |
Each video's list below names every ringed bird that could appear in it. Not every listed
bird appears.
| video |
candidate birds |
feeder_1.mp4 |
arlb, cgba, cyap, gagr, gamg, gaob, garw, mamo, mays, opma, yaws |
feeder_2.mp4 |
bowa, camc, gamb, mwwa, ooca, oyca, sswa, wpap, wsca, yamp, yaro, ypca |
feeder_3.mp4 |
arsg, bpay, gabs, gacm, gagp, gamm, garr, gaym, gays, gayw, gmwa, lgwa, mapm, mawr, oarl, occa, oowa, ppca, roca, rwas, sbca, ssca, yagg, yass, yoca |
feeder_4.mp4 |
arsg, bpay, brca, bsma, bwya, byca, cawy, gmwa, mawr, ppas, roca, rwas |
Example: a bird coded oaor has orange over aluminium on its own left leg and orange over red
on its right leg. A bird whose rings you cannot match to its video's list (for example an
unringed bird) is ignored.
Task
For every listed bird that feeds in each video, reconstruct every feeding bout as the list
of its pecks.
- Peck: one dip of the head to the food. The bird lowers its head until its bill reaches
a food bundle, then raises it again. One dip is one peck, however long the bill stays at the
food and however many small nibbling or tugging movements it makes there; the next peck
starts only after the head has come back up. Touching the perch, the post or the bird's own
feet is not a peck. A peck's time is the first frame in which the bill touches the food, given in
seconds from the start of that video file (frame index ÷ 25).
- Feeding bout: take all pecks by one bird in one video, in time order, and split them
wherever two consecutive pecks are 20 seconds or more apart. Each resulting group is one
bout; a lone peck is a bout of one. Pecks by different birds never share a bout, and two
birds feeding at the same time each have their own bouts.
Output
Write output/bouts.json:
{
"bouts": [
{"video": "feeder_2.mp4", "bird": "gamb", "peck_times_s": [1500.2, 1501.9, 1503.0, 1506.4]},
{"video": "feeder_4.mp4", "bird": "byca", "peck_times_s": [640.0]}
]
}
- One entry per bout, any order.
video must be one of the four files and bird must be on
that video's list. peck_times_s lists every peck of the bout.
- An entry matches a real bout only if all of these hold: the bird is right; its first and
last peck times are each within 2 seconds of the real bout's first and last pecks; its number
of pecks is within 15% of the real count (at least ±1); and at least 80% of the real pecks
can each be paired with a different listed peck time within 0.3 seconds. A bout split or
merged differently from the definition above does not match.
- Every entry you submit counts: entries that match no real bout count against you, and so
does listing the same bout under several birds.
- Write the file early and keep overwriting it as you go. There is a time limit, and only what
is in
output/bouts.json when the session ends is graded.
Rules
- Work only inside this working directory. Do not read, write or search outside it.
- There is no internet access. Do not rely on memory of these birds, this study or any
published annotations; every entry must come from what you observe in the videos.
- Available tools are what is installed:
ffmpeg/ffprobe, and Python 3 with numpy,
scipy, pandas, pillow, opencv-python-headless and matplotlib. pip install
will not work. You can extract frames or crops and look at them as images.