STREETS: A Novel Camera Network Dataset for Traffic Flow
Authors: Corey Snyder, Minh Do
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our dataset provides over 4 million still images across 2.5 months and one hundred web cameras in suburban Lake County, IL. We present benchmarking tasks and baseline results for one such task to guide how future work may use our dataset. Benchmark evaluation tasks and baseline results using STREETS. We conduct experiments at the camera view depicted in Fig. 1.(c). We predict the vehicle counts on the inbound vertex for each date in 7/15/2019-7/18/2019 and vary the choices of K, L, and δ to examine the importance of the traffic graph. Tables 4 and 5 present results for each model with varying choices of L and K under both MAE and MAEc. |
| Researcher Affiliation | Academia | Corey Snyder University of Illinois cesnyde2@illinois.edu Minh N. Do University of Illinois minhdo@illinois.edu |
| Pseudocode | No | The paper describes algorithms and procedures (e.g., how Mask R-CNN was used, data collection steps) but does not present them in a structured pseudocode or algorithm block. |
| Open Source Code | Yes | Codes and data for this model are available on our accompanying Git Hub2 and dataset, respectively. 2Link to STREETS Git Hub. |
| Open Datasets | Yes | In this paper, we introduce STREETS, a novel traffic flow dataset from publicly available web cameras in the suburbs of Chicago, IL. Link to STREETS dataset. [7] Corey Snyder and Minh Do. Data for streets: A novel camera network dataset for traffic flow, 2019. |
| Dataset Splits | Yes | we retrain the network on a collection of nearly 3,000 labeled images with an 85-15 training-validation split. trained a Res Net-50 [49] model with an 80-20 training-validation split to perform the binary classification. |
| Hardware Specification | No | The paper mentions implementing models (Mask R-CNN, ResNet-50, RFR, SVR, LR, ANN) but does not specify any hardware details such as GPU/CPU models, memory, or specific computing environments used for training or inference. |
| Software Dependencies | No | The RFR, SVR, and LR models are all constructed using their corresponding default models in the scikit-learn Python library. The ANN model is implemented in Py Torch with hidden layers of sizes 200 and 100. The paper mentions software libraries but does not provide specific version numbers for them. |
| Experiment Setup | Yes | Training is performed for 10,000 steps with batch size of 32 using Stochastic Gradient Descent at learning rate 10 3 and momentum 0.9. Traffic data from the weekdays of 6/5/2019-7/12/2019 are used as training data for each model. The ANN model is implemented in Py Torch with hidden layers of sizes 200 and 100. |