Meta's Segment Anything Model (SAM)

The competition focuses on specialization, SAM is more of a generalized model. As of now, there are no guidelines on how to use this as a downstream. Out of curiosity, have you tried it already in the competition?

View post 0

💬 Feedback and suggestion

I guess the leaderboard hyperlinks are invalid. A message on the page says “Leaderboard is not released yet” but I can see the leaderboard.
Checkout this page: AIcrowd | Scene Understanding for Autonomous Drone Delivery (SUADD'23) | Leaderboards
Both the leaderboard links point to mono-depth-perception one.

View post 0

👥 Looking for teammates?

Hey guys,

I am Gaurav Singhal working as a data scientist in health tech. I have been working on image segmentation for 2 years, won the food-image-recognition challenge in 2021, and achieved #2 last year. I do not have a dedicated GPU except for Colab Pro and one shared server with V100.
With this challenge, I am looking to expand my domain knowledge and ofcourse would like to collaborate.

I have experience with CNN and Vision transformer-based methods for segmentation. I would like to improve the method further to establish new state-of-the-art in the process.

Here is my LinkedIn https://www.linkedin.com/in/gaurav-singhal93/. Ping me if you would like to collaborate.

View post 1

💡 Solutions are Public

Nice approach by Lab_3i. QueryInst with Swin.

View post 0

End of Round 2⏱️

Thanks, @Mykola_Lavreniuk. I hope our techniques are different from each other to broaden our knowledge horizons. I would love to collaborate with you and your team if I will be invited to co-author.
Congratulation to @Lab_zi and @Camaro

View post 2

End of Round 2⏱️

Great work guys.
Thank you @mohanty, @shivam for making this challenge happen.
Thanks everyone on the leaderboard, it was really fun this time. In the previous week it felt like a roller coaster ride with all the score shuffling. I expect the final leaderboard will see a very small variance in the score and the results don’t shuffle by a huge gap.
Congratulations to @team_zi, and great work @saidinesh_pola, @nivedita_rufus and @unnikrishnan.r

View post 5

Getting less mAP on MMdetection

I guess I know what top participants are using but I cannot reveal that at the moment since the competition is still ongoing. What I can tell you is:

• I am highly confident that they are using MMDetection
• You haven’t tried the previous challenge’s best solution. I know that because I won it, and as simple as Mask RCNN yielded slightly better performance than what you are getting now
• Data augmentation, Multi-scale training do make a significant improvement
• The hyper-paramters which are well known in image classification tasks play a role in instance segmentation but not too much, the research has its own hyper-parameters that you will find in test_config
• Regarding ensemble. Yes, it is not working in this challenge because AP@50 is quite robust and naive ensembling will not be enough. You need some more post-processing steps to filter out false positives.

View post 3

Local Run Produces Different AP

My theory:

1. Validation results that you see in Gitlab are just a model sanity check. I think it has nothing to do with what you see on your local machine. It just checks if your submission is worthy to assign pods to evaluate.
2. Your local validation results are overfitted for the very obvious reason that some ~950 images are the same as that of the train. In case you have already removed these you will run into the problem of class imbalance with some classes not available at all. I created a new validation set for myself and I can say that it’s the best, the result I see on my local machine gives ~+6.0% (positive variance) jump on a test score.

View post 3

[Announcement] Leaderboard Winners

Congratulations to all the winners.

View post 1

When will release round 2 baseline?

Regarding baseline, I guess the AICrowd team is working on it and it is will be out soon, (it’s expected today). However, you can start your R&D with this notebook AIcrowd | [First Baseline + Explainer] Getting Started With ROUND 2 | Posts.
Regarding live score, I think it’s live but submissions are failing a lot lately because of time-out issues (at least for me).

View post 1

:aicrowd: [Update] Round 2 of Data Purchasing Challenge is now live!

Efficient-b0 tends to learn faster compared to its family members. Since the dataset is smaller, with 64 as batch size, unfreezing all layers would be better for b0 compared to the same config in b4, the quality of purchase depends on it, I think you may perform a small experiment with the best-unlabelled images (64 batches, 0.x LR, y epochs) and see if b0 outperforms b4.
In any case, I don’t think it matters. If all the evaluations will run on the same configuration then the performance will be equally good or bad for all the participants. However, with the above experiment, I guess you and we would be able to see if the purchase makes any sense or not.

View post 0

Need Clarification for Round 2

tfreidel raised the same bug here (:aicrowd: [Update] Round 2 of Data Purchasing Challenge is now live! - #11 by tfriedel). I also think it should be aggregated_dataset instead of training_dataset . Although it is in local_evaluation.py which will not be part of any sort of evaluation.

View post 0

[Resolution] Bugs With Getting Started Of Round 2

Yep. Thanks.
Could you add extract, rename script in GitLab repo, or maybe change the local_evaluation.py just like you did in colab.

View post 1

[Resolution] Bugs With Getting Started Of Round 2

This post just focuses on the Magic box part. I haven’t checked out the methods yet, I hope there’s nothing left out there but I’ll check and report any inconsistencies.
You are right, the notebook uses the public_ prefix in dataset declarations but the GitLab code doesn’t. In any case, the spelling is still messed up, no big of an issue but maybe you want to correct it.

View post 0

[Resolution] Bugs With Getting Started Of Round 2

Some issues that I faced with Getting started of Round 2.

• Dataset download with prefix public_*, however local_evaluation.py uses directory without public_*
• Spelling mistake for unlabelled dataset, currently it is public_unlabeled.zip rather it should be public_unlabelled.zip. You will see this once you download the dataset, not while listing it.

I may be wrong, if so please correct me @shivam @mohanty.

I have put together a Magic box (based on magic box from Round 1) that will make things easy and make the repository ready to use. Here are a few actions that I am trying to achieve.

• Cloning the repository for Round 2
• Downloading datasets for Round 2 and putting them in relevant directories (abiding the latest local_evaluation.py file)
• Renaming the dataset directories as per latest local_evaluation.py

Magic Box for Colab

try:
  import os
  if first_run and os.path.exists("/content/data-purchasing-challenge-2022-starter-kit/data/training"):
    first_run = False
except:
  first_run = True

if first_run:
  %cd /content/
  !git clone http://gitlab.aicrowd.com/zew/data-purchasing-challenge-2022-starter-kit.git > /dev/null
  %cd data-purchasing-challenge-2022-starter-kit
  !aicrowd dataset list -c data-purchasing-challenge-2022 | grep -e 'v0.2'
  !aicrowd dataset download -c data-purchasing-challenge-2022 *-v0.2-rc4.zip
  !mkdir -p data/
  !mkdir -p data/v0.2-rc4
  !mv *.zip data/v0.2-rc4 && cd data/v0.2-rc4 && echo "Extracting dataset" && ls *.zip | xargs -n1 -I{} bash -c "unzip \*.zip > /dev/null"
  !mv data/v0.2-rc4/public_debug data/v0.2-rc4/debug
  !mv data/v0.2-rc4/public_training data/v0.2-rc4/training
  !mv data/v0.2-rc4/public_unlabeled data/v0.2-rc4/unlabelled
  !mv data/v0.2-rc4/public_validation data/v0.2-rc4/validation

Magic Box for Local System

#!/bin/bash

git clone http://gitlab.aicrowd.com/zew/data-purchasing-challenge-2022-starter-kit.git
cd data-purchasing-challenge-2022-starter-kit
aicrowd dataset list -c data-purchasing-challenge-2022 | grep -e 'v0.2'
aicrowd dataset download -c data-purchasing-challenge-2022 *-v0.2-rc4.zip
mkdir -p data/
mkdir -p data/v0.2-rc4
mv *.zip data/v0.2-rc4 && cd data/v0.2-rc4 && echo "Extracting dataset" && ls *.zip | xargs -n1 -I{} bash -c "unzip \*.zip > /dev/null"
mv public_debug debug
mv public_training training
mv public_unlabeled unlabelled
mv public_validation validation

Put the above code in magic_box.sh and execute
>>> bash magic_box.sh

Please do let me know any improvements or questions in the comments below, I would be glad to help you.
Click on if this post was of any help

View post 3

0.9+ Baseline Solution for Part 1 of Challenge

Buying low-accuracy labels (dents) make the most sense in this challenge, sounds easy but challenging. I have the exact same heuristic with the addition of one more policy (don’t judge by my score it is only a baseline, I didn’t submit the solution because I had too much on my plate ).
Just to give a perspective, here is the confusion matrix of dent_small and dent_large respectively.

[[7327  392]
 [ 719 1562]]

[[8736  116]
 [ 328  820]]

The sequence is - tn, fp, fn, tp. fp+fn for dent classes is much more than those of scratch class.
One of the approaches to deal with this can be the weighted loss function. I haven’t implemented it so can’t tell the improvement.

View post 0

Error : no gpu

Your code must be supported for CUDA + Put gpu:true in aicrowd.json
If you still have the problem then only @shivam can help you.

View post 0

Brainstorming On Augmentations

Regarding 4, I didn’t want to make any comment on your coding skills. It’s good to follow coding good practices, always useful, and makes the code reusable, understandable, etc. At least for me, I try to write code that should not require not much to make it production-ready.
I didn’t mean any offense.

View post 1

Brainstorming On Augmentations

I tried your code and nobody asked but here are my 2 cents:

1. I don’t understand (if anybody does then please help) why you have written separate dataset classes. The dataset classes are self-sufficient on their own and are meant to be used the way they were created.
2. You have completely neglected the pre-training phase. You are doing it in the purchase phase which is a different purpose altogether.
3. I tried your augmentations and hyper-parameters but wasn’t able to reproduce the results. I am using the provided dataset class and pre-training phase, not the way you have done it. Maybe this could be a reason why I am not able to reproduce the results.
4. Sorry to say this but the code is really messy.

View post 0

Learn How To Make First Baseline Model With 0.44+ Accuracy on LeaderBoard [🎥 Tutorial]

Thanks for the feedback. I have made the relevant changes.

View post 0