Unsupervised Object Keypoint Learning using Local Spatial Predictability
Authors: Anand Gopalakrishnan, Sjoerd van Steenkiste, Jürgen Schmidhuber
ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the efficacy of Perma Key on Atari where it learns keypoints corresponding to the most salient object parts and is robust to certain visual distractors. Further, on downstream RL tasks in the Atari domain we demonstrate how agents equipped with our keypoints outperform those using competing alternatives, even on challenging environments with moving backgrounds or distractor objects. |
| Researcher Affiliation | Academia | Anand Gopalakrishnan, Sjoerd van Steenkiste, J urgen Schmidhuber The Swiss AI Lab IDSIA, USI, SUPSI {anand, sjoerd, juergen}@idsia.ch |
| Pseudocode | No | The paper describes the methods in text and diagrams, but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Code to reproduce all experiments is available at https://github.com/agopal42/permakey. |
| Open Datasets | Yes | To obtain Atari game frames, we use rollouts of various pre-trained agents in the Atari Model Zoo (Such et al., 2019). |
| Dataset Splits | Yes | We split the aggregated set of game frames for each of the chosen environments into separate train, validation and test sets of 85,000, 5000 and 5000 samples respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. It mentions 'available amount of compute' generally. |
| Software Dependencies | No | The paper mentions using the 'Adam optimizer (Kingma & Ba, 2015)' but does not provide specific software dependency versions (e.g., library or framework names with version numbers like Python 3.x, PyTorch 1.x) needed to replicate the experiment environment. |
| Experiment Setup | Yes | We train our method using the Adam optimizer (Kingma & Ba, 2015) with a initial learning rate of 0.0002 and decay rate of 0.85 every 10000 steps. We use a batch size of 32 and in all cases train for 100 epochs, using use early stopping of 10 epochs on the validation set to prevent overfitting. |