#vesuvius-challenge-surface-detection

1 messages · Page 1 of 1 (latest)

charred zephyr
#

Hi everyone! Just joined the competition today, excited to see what people come up with in the end 😄

neon copper
#

just a hobbyist here, pleased just to have successfully trained a model, made a couple of submissions, and learned something along the way. happy to bounce ideas around if anyone needs sounding board. I posted one of mine on the kaggle forum. Fun challenge and can't wait until we're able to unroll and read these scrolls!

shut moatBOT
#
jordi45698 has been warned

Reason: Posted an invite

sharp timber
#

Welcome to the Vesuvius Challenge Surface Detection competition. I am the competition host. Feel free to ask questions regarding the contest.

spiral oyster
#

Hi
I have taken part in this competition and hoping to form a team
I have good theoretical knowledge and some practical as well. Want to team up if someone is interested please dm

sharp timber
#

Bonus prizes available! Share your code!!! More info here

iron star
#

https://media.discordapp.net/attachments/1444971360047726605/1445085758598938824/image1.gif?ex=692f107d&is=692dbefd&hm=94f18cd6e7350e7cc612826beb5d11a9fd125485a58ee1e39a16a03b6f9e2426&=&width=237&height=315
https://media.discordapp.net/attachments/1444971360047726605/1445085766937088000/image2.gif?ex=692f107f&is=692dbeff&hm=51e8429e6818b166e21485a613e8f0c706d64c765aefc93f65a7bcefa10907c2&=&width=864&height=1152
https://media.discordapp.net/attachments/1444971360047726605/1445085774562197535/image3.gif?ex=692f1081&is=692dbf01&hm=e520e8e4edd4eea02e82168a7059a868ea59c19d9b90c7c34402f7bb3616c76f&=&width=864&height=1152
https://media.discordapp.net/attachments/1444971360047726605/1445085781801566319/image4.gif?ex=692f1082&is=692dbf02&hm=bdc0715977fdcda4b7804916e5bfb36af1d3132f535d1b4327894a067fbfc769&=&width=725&height=907

charred zephyr
#

Just posted in looking for a team but I'll post here too. I'd love to participate in this competition but don't have the time bandwidth to do it solo. If anyone wants a part time software engineer, please ping me!

charred zephyr
#

I have a question about this particular challenge. Is it 9hrs run time only, or 9hrs including training?

#

Also, is there somewhere we can get our hands on more scans to better train our models?

#

Or is the challenge only to do the best we can with what is provided in this challenge itself?

charred zephyr
#

Oh one more, what is the actual goal of the competition? I thought it was something like providing an unwrappable 3d mesh/voxel of detected surface(s) or 2d images of the surface fragments or something like that, but the example notebook just outputs prediction data from a CSV. What does a winning submission look like?

charred zephyr
#

Hmm doesn't look like I can share an image, but I got my first result out of it! 🙂

charred zephyr
#

Is it ok to ditch the labels in the data altogether? What if I want my code to learn how to label data on its own?

sharp timber
#

What do you mean by ditch?

charred zephyr
#

Does the model need to be trained on the masks, or can it be trained on just the voxel data?

sharp timber
charred zephyr
#

Right, but I can output masks programmatically from 3D, no need for machine learning for that. 😁

glossy osprey
zealous fulcrum
#

On the new dataset, label images having all pixels are with value 2. does anyone else having similar issue?

df_train = pd.read_csv('/kaggle/input/vesuvius-challenge-surface-detection/train.csv')
df_test = pd.read_csv('/kaggle/input/vesuvius-challenge-surface-detection/test.csv')
directory_path = Path('/kaggle/input/vesuvius-challenge-surface-detection/deprecated_train_images') # '.' refers to the current directory
deprecated_ids = [int(p.with_suffix("").name) for p in directory_path.iterdir() if p.is_file()]
df_train = df_train[~df_train["id"].isin(deprecated_ids)]
df_train.reset_index(drop=True, inplace=True)
df_test = df_test[~df_test["id"].isin(deprecated_ids)]
df_test.reset_index(drop=True, inplace=True)
for ii in df_train['id']:
img = Image.open(f"/kaggle/input/vesuvius-challenge-surface-detection/train_labels/{ii}.tif").convert("L")
a = np.array(img, dtype=np.float32)
print(ii,a.min(),a.max())

2290837 2.0 2.0
3320274 2.0 2.0
8862040 2.0 2.0
11460685 2.0 2.0
11630450 2.0 2.0
15307632 2.0 2.0
17283971 2.0 2.0
19797301 2.0 2.0
26894125 2.0 2.0
29754811 2.0 2.0
38034250 2.0 2.0
40625686 2.0 2.0
44701864 2.0 2.0
45525309 2.0 2.0

sharp timber
dire mauve
#

Looking for help from fellow Kagglers:
I have performed train / inference on local environment with python 3.10 / torch 2.8
I cannot reproduce my results with Kaggle Notebook as they run on python 3.12 (so having the pip .whl uploaded as a dataset does not work either)
Because it is a notebook competition, I cannot just submit the zip. Any idea on how to tackle this issue ?

thick peak
#

Load the model and make inference only. Predictions should be deterministic.

dire mauve
thick peak
#

weights are weigths, no matter the version

idle gust
#

Is anyone else facing a lot of kaggle errors/submission errors on submission? On some instances resubmission works but often I end up with another round of kaggle error and run out of submissions for the day. I've verified my submission format and it appears to be in line with instructions as well

brazen lotus
#

Im having the same problem. I can't figure out why all my submission for this challenge are failing.

idle gust
brazen lotus
brave socket
#

just starting this challenge, so to start which is the basic book template as i learn the kaggle competition..

wise kelp
#

hHello everyone, I'm having problems with inference performance on the kaggle platform. I was trying to submit a U-Net to a challenge and, the first time (some days ago) each inference was taking around 30 seconds to complete. Now it is taking more than 2 minutes and 30 seconds. Do you have/had the same problem? Thanks!

storm citrus