update readme

examples/bdd_oia/README.md

@@ -2,7 +2,7 @@
 
 This example shows an implementation of [BDD-OIA](https://twizwei.github.io/bddoia_project/) task. The BDD-OIA dataset comprises frames extracted from driving scene videos, which are utilized for autonomous driving predictions. Each frame is annotated with 4 binary labels, indicating the possible actions, namely $\textsf{move forward}$, $\textsf{stop}$, $\textsf{turn left}$, $\textsf{turn right}$. Each frame is also annotated with 21 intermediate binary concepts such as $\textsf{red light}$, $\textsf{road clear}$, etc., underlying the reasons for the possible actions.
 
-The objective is to predict possible actions for each frame. During training, we only make use of the label supervision, along with a knowledge base, which comprises information about the relations between concepts and actions, e.g., $\textsf{red light} \lor \textsf{traffic sign} \lor \textsf{obstacle} \implies \textsf{stop}$.
+The objective is to predict possible actions for each frame. During training, we only make use of the label supervision, along with a knowledge base, which comprises information about the relations between concepts and actions, e.g., $\textsf{red light} \lor \textsf{traffic sign} \lor \textsf{obstacle} \implies \textsf{stop}$. The training set consists of 16,000 frames, while the test set contains 4,500 annotated data points.
 
 Before usage, the dataset was pre-processed by [Marconato et al. (2023)](https://proceedings.neurips.cc/paper_files/paper/2023/file/e560202b6e779a82478edb46c6f8f4dd-Paper-Conference.pdf) using a pretrained Faster-RCNN model on BDD-100k, in conjunction with the first module in CBM-AUC [(Sawada & Nakamura, 2022)](https://arxiv.org/abs/2202.01459), resulting in embeddings of dimension 2048.
 
@@ -10,7 +10,8 @@ Before usage, the dataset was pre-processed by [Marconato et al. (2023)](https:/
 
 ```bash
 pip install -r requirements.txt
-cd dataset & unzip dataset.zip
+cd dataset
+unzip dataset.zip
 cd ..
 python main.py
 ```