From 82158c3cc5208f40611675d9744e5d9a98328ceb Mon Sep 17 00:00:00 2001 From: jshen12 Date: Fri, 22 Mar 2024 20:13:15 -0700 Subject: [PATCH] image fix --- _posts/2024-03-21-team19-trajectory-prediction.md | 2 +- assets/images/19/{image.png => efficiency.png} | Bin 2 files changed, 1 insertion(+), 1 deletion(-) rename assets/images/19/{image.png => efficiency.png} (100%) diff --git a/_posts/2024-03-21-team19-trajectory-prediction.md b/_posts/2024-03-21-team19-trajectory-prediction.md index 07b9541..3c7f3b0 100644 --- a/_posts/2024-03-21-team19-trajectory-prediction.md +++ b/_posts/2024-03-21-team19-trajectory-prediction.md @@ -123,7 +123,7 @@ Where $$L_{\text{traj}}$$ represents the negative Gaussian log-likelihood for th The trajectory prediction capabilities of autonomous vehicle (AV) systems are benchmarked against metrics that reflect their precision and efficiency in real-world scenarios. VectorNet stands out in this domain, as illustrated by both its ADE (Average Displacement Error) and DE@3s metrics, which demonstrate its exceptional ability to predict the future positions of on-road agents. It boasts an ADE of 1.81 meters, showcasing superior average accuracy across time steps, and a DE@3s of 4.01 meters, highlighting its precision in short-term trajectory forecasting. This performance surpasses traditional approaches like constant velocity models and LSTM-based architectures, which were once standard. -![Vectornet efficiency]({{ 'assets/images/19/performance.png' | relative_url }}) +![Vectornet efficiency]({{ 'assets/images/19/efficiency.png' | relative_url }}) {: style="width: 800px; max-width: 100%;"} *Fig 4. FLOPs and param # for Vectornet compared with other Resnet models [4]*. diff --git a/assets/images/19/image.png b/assets/images/19/efficiency.png similarity index 100% rename from assets/images/19/image.png rename to assets/images/19/efficiency.png