#birdclef-2026

1 messages · Page 1 of 1 (latest)

winged kraken
#

Hi, I'm looking for teammates for this competition.

plush tiger
winged kraken
calm shore
#

Hi,
I am an AI engineer with over 2 years of experience in this domain. I am looking for a team or teammate for this competition.
Interested person, please dm me.

hardy osprey
#

feedback appreciatedhttps://www.kaggle.com/code/suhanigupta04/birdclef-2026

crisp sparrow
#

Hello everyone, I am looking for a mentor or collaborator for this competition. If anybody is looking for a team mate then ping me. I am beginner who has built some basic machine learning models.

warm wadi
#

"Hi everyone! I’m a complete beginner to Kaggle and Bioacoustics. I have zero prior experience with these algorithms, but I have a lot of time to dedicate to this project. I'm looking for a mentor-style team where I can help with data cleaning, researching species behavior, or running experiments while I learn the technical side. I'm honest about my starting point and just want to contribute however I can!

livid gulch
#

Hi, I have a couple of questions about what's allowed in this competition:

  1. Is it allowed to query an external API like pygbif during the notebook runtime to retrieve the full taxonomical hierarchy (class, order, family, genus) for the target species, and use that structure as part of the model architecture?

  2. Is it allowed to use more than one model serving different purposes within the same notebook? For example, a first model to filter out unclassifiable clips, and a second model to actually classify the remaining ones.

Thanks!

kindred bear
hexed sigil
#

I think there is two main approaches up to this point.

  1. Using pretrained Google Perch and fine-tuning/expanding with this competition.
  2. Using a pretrained Image based CNN backbone (Class BirdModel) to extract features and train a classifier on top.

The most voted code is blending this two methods, but I don't yet understand how or why. Am I wrong? I also saw someone using attention heads, but this might be too complicated for me.
What approach would you recommend for a relative beginner?

kindred bear
#

Just out of curiosity, how long are you usually waiting for your score when submitting the inference notebook? I'm wondering whether I should try to implement a faster approach.

valid quest
#

hey, has anyone compared the performance of a CNN vs Perch, which one is better?

candid dune
#

Have one advice.
Kaggle already have birdclef in 2021, 2022, 2023, 2024, 2025, 2026 so you can see notebooks and take some hints from it. Wish you success guys!

If you want to know birdclef is only for birds, but birdclef + for birds and other animals.

green peak
green peak
#

Actually, does anyone know how many data points are in the hidden test set?

kindred bear
#

I had that problem too until I realised

kindred bear
#

But the public leaderboard is calculated on 34% so I’m not sure whether the whole test data has 600, or did they just mean the public leaderboard amount

green peak
kindred bear
#

i had the same problem, my notebook would get timeout after 5-10min

green peak
#

Ah sick. That makes sense. Cheers. I probably should have read the rules a bit better 😂

quiet tulip
#

check my bio 😁

forest sonnet
#

check my bio 😁

valid quest
#

Hi, I have a 0.930 pipeline, Looking for team mates for this competition.

heady stone
candid atlas
#

Hi everyone, can someone help me, I tried submitting my notebook, the notebook itself runs successfully but the scoring times out, I turned the internet off and I'm not using any GPU. What might be the issue?

marble sentinel
candid atlas
#

When I run it on Kaggle, it takes about 40 minutes

marble sentinel
valid quest
hearty ferry
#

Check my bio 😁

candid atlas
#

Thanks I will try your suggestions!

patent horizon
#

Hello,

My notebook is perfectly running and predicting the labels of the training set but when I submit it does not find or detect any test files and the test directory is always empty.

Can you please help?

valid quest
patent horizon
valid quest
patent horizon
#

I understand what you are saying but when I add a debugging print statement to check all the files in the submission and check it from the logs, I have the following:

134.4s 270 Found 1 files/directories in /kaggle/input/competitions/birdclef-2026/test_soundscapes:
134.4s 271 - readme.txt

so where to find the 600 .ogg files? or what path to use in the submitted notebook to have the prediction generated?

marble sentinel
#

There are 2 stages to submit a notebook:

  • Save and run all: at this stage the notebook sees no .ogg files in the test_soundscapes folder. Usually there are some dummy files, but for this competition there didn't put any. So to work around that, most public notebooks detect that there are no files and use a few train_soundscapes instead
  • Submit to competition: the notebook is run again, this time with 600 .ogg files in the test_soundscapes folder. It needs to use those files and generate a submission file from them.

You can check some public notebook to see how they handle this quirk.

spark monolith
#

Hello everyone .I looking for team to be a part of competitions .. But I'm begginar . I came here to learn new things. I have good patience to be in teamwork. How to be part of competitions please anyone help me ..

kindred night
#

Hello, is there any Italian who is attending this competition?

patent horizon
cloud pumice
#

🚨 STOP sending boring CVs.

If you want your dream career, you need to stop using basic AI and start using the STAR Method.

I made a quick reel showing:
✅ How to turn a "Boring CV" into a "Selected" CV.
✅ The prompt that Senior Recruiters actually love.
✅ The #1 mistake that gets you rejected immediately.

Get the prompt here:
https://www.instagram.com/reel/DYOrbw3zcxI/?igsh=MTAzdTg0dzQ1dTFnbg==

broken sinew
#

Anyone struggling in Consistency and want to learn together.
DM me.

candid atlas
#

Hey guys, does anyone know if mixup helps with improving the performance of the model, I am using Perch + MLP + Late fusion with perch meta. I wanted to add Mixup, but it hurt the LB. In the discussions, I am getting mixed feedbacks, does anyone know how and when to use it?

pseudo nexus