You Mostly Walk Alone: Analyzing Feature Attribution in Trajectory Prediction
Authors: Osama Makansi, Julius Von Kügelgen, Francesco Locatello, Peter Vincent Gehler, Dominik Janzing, Thomas Brox, Bernhard Schölkopf
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Applying this procedure to stateof-the-art trajectory prediction methods on standard benchmark datasets shows that they are, in fact, unable to reason about interactions. ...Overall, our analysis shows that established trajectory prediction datasets are suboptimal for benchmarking the learning of interactions among agents, but existing approaches do have the capability to learn such interactions on more appropriate datasets. |
| Researcher Affiliation | Collaboration | Osama Makansi 2, Julius von K ugelgen3,4, Francesco Locatello1, Peter Gehler1, Dominik Janzing1, Thomas Brox*1,2, and Bernhard Sch olkopf*1,3 1Amazon, 2University of Freiburg, 3Max Planck Institute for Intelligent Systems T ubingen, 4University of Cambridge |
| Pseudocode | No | The paper describes the proposed method and analysis techniques in prose and with figures, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not explicitly state that its source code is released or provide a link to a code repository. |
| Open Datasets | Yes | ETH-UCY (Pellegrini et al., 2009; Leal-Taix e et al., 2014), SDD (Robicquet et al., 2016), and nu Scenes (Caesar et al., 2020)... Sport VU (Yue et al., 2014) is a tracking dataset for games recorded from multiple seasons of the NBA. |
| Dataset Splits | Yes | ETH-UCY is one of the most common benchmarks for trajectory prediction. ...We use the standard 5-fold cross validation for our analysis. ...SDD... We follow the standard train/test split used in previous works (Mangalam et al., 2020; Gupta et al., 2018). ...nu Scenes... We also use the standard training/testing splits (Salzmann et al., 2020). ...Sport VU... We pre-process the dataset to remove short scenes and randomly select 7,000 scenes for training and 100 scenes for testing. |
| Hardware Specification | No | The paper mentions 'Amazon Web Services for providing enough resources to conduct the experiments' but does not specify any particular hardware models (GPU, CPU, or specific AWS instance types). |
| Software Dependencies | No | The paper mentions using LSTMs, GANs, conditional VAEs, Graph Neural Networks, and Transformers, but it does not specify software dependencies with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x, Python 3.x). |
| Experiment Setup | No | The paper discusses model architectures and data processing, and mentions making the history encoder 'one layer deeper' for Trajectron++Edge, but it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) or system-level training configurations for the experiments conducted. |