IROS 2019 Lifelong Robotic Vision Challenge: Lifelong Object Recognition

Organized by chuanlinlan - Current server time: Oct. 27, 2020, 12:33 p.m. UTC

Previous

First phase
July 17, 2019, midnight UTC

Current

Final Phase
Oct. 8, 2019, 11:59 a.m. UTC

End

Competition Ends
Never

Lifelong Object Recognition

  • This challenge intends to explore how to leverage the knowledge learned from previous tasks that could generalize to new task effectively, and also how to efficiently memorize of previously learned tasks. Making the robot behaves like the human with knowledge transfer, association, and combination capabilities.

  • To our best knowledge, the provided lifelong object recognition dataset is the 1st one that explicitly indicates the task difficulty under the incremental setting, which is able to foster the lifelong/continual/incremental learning in a supervised/semi-supervised manner. Different from previous instance/class-incremental task, the difficulty-incremental learning is to test the model’s capability over continuous learning when faced with multiple environmental factors, such as illumination, occlusion, camera-object distances/angles, clutter, and context information in both low and high dynamic scenes.

Task-specific Rules

  • The methods should be incremental, which means the model should be only trained over the current task, and test over all previous, current, future tasks. In the 1st round we provide 9 batches of datasets, for each batch, we have train/validation/test data splits. The core of this incremental learning setting is, we need to train on the 1st batch of the dataset, and then 2nd batch, 3rd batch, until the 9th batch, and use the learned models after the final batch to obtain the test accuracy of all tasks (batches). The training/validation datasets can only be accessed during the model optimizations, and any participant use the testing dataset once detected will be removed from the rank list (after the 1st round, the top-ranked participants should provide reproducible procedures).
  • We hold our competition on the Codalab website, and the participants should submit their prediction results (object label) and evaluate their predictions online with our ground truth.
  • The participants who achieve the high ranking results are encouraged to deliver an oral presentation in IROS 2019 competition section and have an official award from IROS onsite.

OpenLORIS-Object Dataset

Evaluation Procedure

  • The results you submitted are required to be named of "test_batch1.csv", "test_batch2.csv" … "test_batch9.csv", and the format of each CSV file is shown below.
  •              file         label_predict
    0 0000 xxx
    1 0001 xxx
    ... ... ...
     

Baseline Model

  • To make every participant quickly familiar with the procedure, we have provided 3 pre-trained models over the provided datasets. Please follow the readme for running the models.
  • To run these models, you are expected to have installed the following libs.
    python 3.7
    numpy 1.16.2
    pandas 0.24.2
    pytorch 1.1.0
    torchvision 0.2.2
    PIL 6.0.0

Important Date

  • The 1st round: 15th July - 30th September
  • The 2nd round: 1st October - 4th November

Discussion Channel

  • You can find more information about our "OpenLORIS-object" dataset on our "lifelong robotic vision" dataset page.
  • If you have any questions about the dataset, pre-trained modules, or any other competition problem please open issues on our github . You are welcome to join the online discussion with our organizers.
  • If you have any problems with your result submission, please contact Chuanlin.
  • QQ Group: 600529306

FAQ

  • Do we have to train our model sequentially as your provided 9 batches ?

Yes. To obtain the consistent and comparable results, you need to train your model and design your own learning algorithm that can train sequentially over provided 9 batches.

  • Can you provide a briefer introduction on how to train the model that can meet the competition requirements ?

For example, when you train the 1st batch of datasets, you can only have access to train/validation data of 1st batch, but need to test over 1-9 batch test datasets; next when you train the model over the 2nd batch of dataset, you can only have access to train/validation data of 2nd batch, but you can also keep some of validation data from 1st batch for learning current model. That means the validation dataset can be kept for the sequential learning task.

  • The motivation of these training procedure constraints ?

This scenario is quite common when the system is deployed as the real-world application, which should be able to update the model day after day, also the system memory is also valuable that we can not keep old dataset on the current systems, sometimes, we only select some coresets of the dataset that we have used. It is non-trivial to pick out the data that can be the summarizations of the encountered data. We constrain our learning/training procedure that approach this kind of real-world scenario.

  • Can we submit the provided baseline models for joining the competition ?

Yes, you can submit the provided baseline models, but as we have tested, the results are not state-of-the-art methods. The models provided can be used for your quicker familiar with the procedure.

  • Is there any state-of-the-art methods that can learn continuous learning strategy ?

Yes, we recommend you go through the recent review paper over this research topic: Continual Lifelong Learning with Neural Networks: A Review.

License

  • OpenLORIS-object dataset is maintained by Qi She

First phase

Start: July 17, 2019, midnight

Final Phase

Start: Oct. 8, 2019, 11:59 a.m.

Description: Final Phase

Competition Ends

Never

You must be logged in to participate in competitions.

Sign In