Learning About Objects by Learning to Interact with Them
Authors: Martin Lohmann, Jordi Salvador, Aniruddha Kembhavi, Roozbeh Mottaghi
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments reveal that this agent learns efficiently and effectively; not just for objects it has interacted with before, but also for novel instances from seen categories as well as novel object categories. and 5 Experiments and Table 1: Object segmentation. Results are for Novel Objects / Novel Spaces scenarios. |
| Researcher Affiliation | Collaboration | Martin Lohmann1, Jordi Salvador1, Aniruddha Kembhavi1,2, Roozbeh Mottaghi1,2 1PRIOR @ Allen Institute for AI 2University of Washington |
| Pseudocode | No | The paper describes algorithms in narrative text (e.g., 'At inference, the model predicts points and force magnitudes... via the following algorithm.') but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides a project page link (https://prior.allenai.org/projects/learning_from_interaction) but does not explicitly state that the source code for the methodology is openly available or provide a direct repository link. |
| Open Datasets | Yes | We train and evaluate our agent within the AI2THOR [30] environment, a near photo-realistic virtual indoor environment of 120 rooms such as kitchens and living rooms with more than 2,000 object instances across 125 categories. and [30] Eric Kolve, Roozbeh Mottaghi, Winson Han, Eli Vander Bilt, Luca Weihs, Alvaro Herrasti, Daniel Gordon, Yuke Zhu, Abhinav Gupta, and Ali Farhadi. AI2-THOR: An Interactive 3D Environment for Visual AI. ar Xiv, 2017. |
| Dataset Splits | No | The paper specifies training and test splits for different scenarios (e.g., 'We use 80 scenes for training... and report results on 20 different rooms' for Novel Spaces), but does not explicitly mention or detail a separate validation dataset split. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models, memory amounts, or detailed computer specifications used for running its experiments. |
| Software Dependencies | No | The paper mentions using AI2THOR, a Unity physics engine, and Mask-RCNN, but it does not specify exact version numbers for any software dependencies, programming languages, or libraries used in the experiments. |
| Experiment Setup | Yes | Our model design is a convolutional neural network inspired by past works in clustering based instance segmentation [42, 16]. As shown in Figure 2, it inputs a single 300 300 RGB+D image and passes it through a UNet style backbone [50]... and In practice, we quantize force magnitudes into 3 bins... and We use a fixed size memory bank (20,000 points). and For self-supervision, we fix the set of forces to f 0 = 5N, f 1 = 30N and f 2 = 200N |