scene cb3ef7b7ef124983b336c781f134cdfb · frame cfd47376746245e899d2223da787fc3d
c1c2c3Six cameras around the car: front-left, front, front-right (top row); back-left, back, back-right (bottom row). Tap the image to open it full size. Pink tags mark the ground-truth objects referenced in the questions.
Reading object references:<c1,CAM_BACK,1159.2,495.0> means object c1 in the back camera at pixel (1159.2, 495.0) of the original 1600×900 image. Models must produce these references themselves, so wrong coordinates mean the model did not find the object.
Upper bound: our fine-tuned model, tested with ground-truth object markers and their coordinate labels drawn on the images (no retraining). Not deployable; measures the grounding bottleneck.
Released checkpoint of InternVL2-2B, referenced as Mini-InternVL2-2B-DA by OpenGVLab. Fully fine-tuned on the original DriveLM split (~378k questions, ~10× our data); under 0.9% of its training data comes from our test scenes.
General-purpose vision-language model, no driving fine-tuning.
OpenGVLab/llama_adapter_v2_multimodal7b
Score = DriveLM final score (0–1, higher is better) on our test split of 3,340 questions. It combines answer accuracy, a GPT judge, language overlap and object matching. Mini-DA is an external checkpoint released by OpenGVLab; all other fine-tuned models are ours.
Showing 3 of 8 models
Q1perception
What are the important objects in the current scene? Those objects will be considered for the future reasoning and driving decision.
Ground truth
There is a walking person to the back of the ego vehicle, a white sedan to the back of the ego vehicle, and a no pedestrian sign to the front of the ego vehicle. The IDs of these objects are <c1,CAM_BACK,1159.2,495.0>, <c2,CAM_BACK,307.5,557.5>, and <c3,CAM_FRONT,1359.5,492.7>.
Best modelours
There is a car to the front of the ego vehicle, a car to the front of the ego vehicle, a car to the front of the ego vehicle, and a car to the front of the ego vehicle. The IDs of these objects are <c1,CAM_FRONT,1080.0,540.0>, <c2,CAM_FRONT,1080.0,540.0>, <c3,CAM_FRONT,1080.0,540.0>, and <c4,CAM_FRONT,1080.0,540.0>.
Oracleours · upper bound
There is a pedestrian riding a motorcycle to the back of the ego vehicle, a white sedan to the back of the ego vehicle, and a pedestrian wearing a black top to the front of the ego vehicle. The IDs of these objects are <c1,CAM_BACK,1159.2,495.0>, <c2,CAM_BACK,307.5,557.5>, and <c3,CAM_FRONT,1359.5,492.7>.
InternVL2-2B (no fine-tuning)zero-shot
In the current scene, the important objects include:
1. **Road**: The primary road is the central focus of the image. It is the main thoroughfare that connects different parts of the area.
2. **Buildings**: There are several buildings visible, including a large office building and a smaller structure with a sign. These buildings are likely important landmarks or facilities.
3. **Trees and Greenery**: There are several trees and greenery along the roadside, which provide a natural element to the urban environment.
4. **Signage**: There is a blue sign with white text on the left side of the road, which likely provides information or directions.
5. **Traffic Signs**: There are traffic signs visible, including a no entry sign and a speed limit sign, which are crucial for traffic management and safety.
These objects are essential for understanding the layout and functionality of the area, as well as for making informed driving decisions.
Fine-tuned 25kours
There is a white sedan to the back of the ego vehicle, a white sedan to the front of the ego vehicle, a white sedan to the front of the ego vehicle, a white sedan to the front of the ego vehicle, and a white sedan to the front of the ego vehicle. The IDs of these objects are <c1,CAM_BACK,1080.0,525.8>, <c2,CAM_FRONT,1080.0,505.8>, <c3,CAM_FRONT,1080.0,505.8>, <c4,CAM_FRONT,1080.0,505.8>, and <c5,CAM_FRONT,1080.0,505.8>.
Fine-tuned 300kours
There is a white sedan to the back of the ego vehicle, a white sedan to the front of the ego vehicle, a white sedan to the back of the ego vehicle, and a white sedan to the front of the ego vehicle. The IDs of these objects are <c1,CAM_BACK,1080.0,520.0>, <c2,CAM_FRONT,1080.0,490.0>, <c3,CAM_BACK,1080.0,520.0>, and <c4,CAM_FRONT,1080.0,490.0>.
OpenGVLab Mini-DAexternal reference
There is a white sedan to the back of the ego vehicle, a pedestrian wearing a blue shirt to the back of the ego vehicle, and a white sedan to the back of the ego vehicle. The IDs of these objects are <0>, <1>, and <2>.
LLaVA-1.6 7Bzero-shot
In the current scene, the important objects for future reasoning and driving decisions include:
1. The road: It's important to be aware of the road's condition, any potential hazards, and the direction of travel.
2. The traffic signs: These provide information about speed limits, road conditions, and directions.
3. The buildings: Understanding the type of buildings and their proximity to the road can help in anticipating pedestrian activity and potential traffic congestion.
4. The trees and vegetation: These can obstruct the view of the road ahead, so it's crucial to be aware of their presence and condition.
5. The parked cars: These can indicate parking restrictions or potential obstacles on the road.
6. The pedestrians: Their presence can affect the speed and direction of travel.
7. The sky: The weather conditions can impact visibility and road conditions.
These objects, along with any other relevant features, will be important for making informed driving decisions.
LLaMA-Adapter v2 7Bzero-shot
In the current scene, the important objects to consider for future reasoning and driving decisions include the parked cars, the traffic lights, and the street signs. The parked cars are located on the side of the road, and their presence might affect the flow of traffic or require extra caution when passing. The traffic lights are crucial for regulating the flow of traffic and ensuring safety at intersections. The street signs provide essential information about the road conditions, directions, and any potential hazards or restrictions. By taking these objects into account, drivers can make informed decisions and navigate the road safely and efficiently.
Q2perceptionmultiple choice
What is the moving status of object <c1,CAM_BACK,1159.2,495.0>? Please select the correct answer from the following options: A. Drive backward. B. Turn right. C. Stopped. D. Going ahead.
Ground truth
D
Best modelours✓ correct
D
Oracleours · upper bound✓ correct
D
InternVL2-2B (no fine-tuning)zero-shotno direct answer
The moving status of object <c1,CAM_BACK,1159.2,495.0> is C. Stopped.
Fine-tuned 25kours✓ correct
D
Fine-tuned 300koursno direct answer
C. Stopped.
OpenGVLab Mini-DAexternal reference✓ correct
D
LLaVA-1.6 7Bzero-shotno direct answer
C. Stopped
LLaMA-Adapter v2 7Bzero-shotno direct answer
The moving status of object <c1,CAM_BACK,1159.2,495.0> is that it is stopped.
Q3prediction
What object should the ego vehicle notice first when the ego vehicle is getting to the next possible location? What is the state of the object that is first noticed by the ego vehicle and what action should the ego vehicle take? What object should the ego vehicle notice second when the ego vehicle is getting to the next possible location? What is the state of the object perceived by the ego vehicle as second and what action should the ego vehicle take? What object should the ego vehicle notice third? What is the state of the object perceived by the ego vehicle as third and what action should the ego vehicle take?
Ground truth
Firstly, notice <c3,CAM_FRONT,1359.5,492.7>. It is a traffic sign, so the ego vehicle should slow down and proceed. Secondly, notice <c1,CAM_BACK,1159.2,495.0>. It is going ahead, so the ego vehicle should continue at the same speed. Thirdly, notice <c2,CAM_BACK,307.5,557.5>. It is also going ahead, so the ego vehicle should continue at the same speed.
Best modelours
Firstly, notice <c1, CAM_FRONT, 1200.0, 500.0>. The object is approaching, so the ego should slow down. Secondly, notice <c2, CAM_FRONT, 1200.0, 500.0>. The object is approaching, so the ego should slow down. Thirdly, notice <c3, CAM_FRONT, 1200.0, 500.0>. The object is approaching, so the ego should slow down.
Oracleours · upper bound
Firstly, notice that <c3,CAM_FRONT,1359.5,492.7>. The object is a traffic sign, so the ego vehicle should continue ahead at the same speed. Secondly, notice that <c2,CAM_BACK,307.5,557.5>. The object is stationary, so the ego vehicle should continue ahead at the same speed. Thirdly, notice that <c1,CAM_BACK,1159.2,495.0>. The object is stationary, so the ego vehicle should continue ahead at the same speed.
InternVL2-2B (no fine-tuning)zero-shot
When the ego vehicle is getting to the next possible location, it should notice the "CAM_FRONT_LEFT" object first. The state of the object perceived by the ego vehicle as the first object is that it is a signpost with a blue background and white text. The action the ego vehicle should take is to proceed straight ahead.
Next, the ego vehicle should notice the "CAM_FRONT_RIGHT" object. The state of the object perceived by the ego vehicle as the second object is that it is a tall building with a reflective glass facade. The action the ego vehicle should take is to continue straight ahead.
The ego vehicle should notice the "CAM_BACK_LEFT" object next. The state of the object perceived by the ego vehicle as the third object is that it is a signpost with a blue background and white text. The action the ego vehicle should take is to proceed straight ahead.
Finally, the ego vehicle should notice the "CAM_BACK_RIGHT" object. The state of the object perceived by the ego vehicle as the fourth object is that it is a tall building with a reflective glass facade. The action the ego vehicle should take is to continue straight ahead.
Fine-tuned 25kours
Firstly, notice that <c3,CAM_FRONT,1080.8,525.0>. The object is going ahead, so the ego vehicle should keep going ahead at the same speed. Secondly, notice that <c2,CAM_FRONT,1080.8,500.8>. The object is going ahead, so the ego vehicle should keep going ahead at the same speed. Thirdly, notice that <c1,CAM_FRONT,1080.8,500.8>. The object is going ahead, so the ego vehicle should keep going ahead at the same speed.
Fine-tuned 300kours
Firstly, notice <c1,CAM_FRONT,1080.0,500.8>. It is going ahead, so the ego vehicle should continue going ahead at the same speed. Secondly, notice <c2,CAM_FRONT,1080.0,500.8>. It is going ahead, so the ego vehicle should continue going ahead at the same speed. Thirdly, notice <c3,CAM_FRONT,1080.0,500.8>. It is going ahead, so the ego vehicle should continue going ahead at the same speed.
OpenGVLab Mini-DAexternal reference
Firstly, notice that the object is going ahead. The ego vehicle should continue going ahead at the same speed. Secondly, notice that the object is going ahead. The ego vehicle should continue going ahead at the same speed. Thirdly, notice that the object is going ahead. The ego vehicle should continue going ahead at the same speed.
LLaVA-1.6 7Bzero-shot
The images provided show a sequence of views from a vehicle's perspective, likely from a camera mounted on the front of the vehicle. Here's a description of the objects and actions the ego vehicle should consider:
1. First object: The vehicle should notice the traffic light first. The traffic light is currently red, indicating that the vehicle should stop.
2. Second object: The vehicle should notice the pedestrian crossing in the next image. The pedestrian crossing is clear, so the vehicle should proceed with caution, checking for any pedestrians who might be crossing.
3. Third object: The vehicle should notice the building in the next image. The building appears to be a commercial or office building, and the vehicle should be aware of the potential for other vehicles entering or exiting the parking lot or driveway.
4. Fourth object: The vehicle should notice the trees in the next image. The trees are lush and green, indicating that the vehicle is in an urban or suburban area with greenery. The vehicle should be cautious of any potential obstacles or debris on the road.
5. Fifth object: The vehicle should notice the road sign in the next image. The sign is not clearly visible, but the vehicle should be prepared to follow any instructions or warnings provided by the sign.
6. Sixth object: The vehicle should notice the parked cars in the next image. The parked cars indicate that the vehicle is approaching a parking area or a busy street with parked vehicles. The vehicle should be prepared to slow down and navigate around the parked cars.
7. Seventh object: The vehicle should notice the building in the next image. The building appears to be a large commercial or office building, and the vehicle should be aware of the potential for other vehicles entering or exiting the parking lot or driveway.
8. Eighth object: The vehicle should notice the trees in the next image. The trees are lush and green, indicating that the vehicle is in an urban or suburban area with greenery. The vehicle should be cautious of any potential obstacles or debris on the road.
9. Ninth object: The vehicle should notice the road sign in the next image. The sign is not clearly visible, but the vehicle should be prepared to follow any instructions or warnings provided by the sign.
10. Tenth object: The vehicle should notice the parked cars in the next image. The parked
LLaMA-Adapter v2 7Bzero-shot
When the ego vehicle is getting to the next possible location, it should first notice the traffic light, which is in the state of "red." The ego vehicle should then take the appropriate action, which is to stop at the red light. Next, the ego vehicle should notice the fire hydrant, which is in the state of "closed." The ego vehicle should then take the appropriate action, which is to proceed cautiously, as it may be necessary to avoid any obstacles or hazards in the area. Finally, the ego vehicle should notice the building, which is in the state of "closed." The ego vehicle should then take the appropriate action, which is to continue driving, as the building is not obstructing the path.
Q4predictionyes / no
Is <c1,CAM_BACK,1159.2,495.0> a traffic sign or a road barrier?
Ground truth
No.
Best modelours✓ correct
No.
Oracleours · upper bound✓ correct
No.
InternVL2-2B (no fine-tuning)zero-shotno direct answer
To determine if the coordinates <c1,CAM_BACK,1159.2,495.0> represent a traffic sign or a road barrier, let's analyze the image in detail:
1. **Location and Orientation**:
- The coordinates are given in the format (x, y, z), where x, y, and z are the coordinates of the point in the image.
- The coordinates are specified as (1159.2, 495.0).
2. **Visual Inspection**:
- **Traffic Sign**: A traffic sign typically has a distinct shape, color, or text that indicates it is meant for drivers to read.
- **Road Barrier**: A road barrier is usually a physical structure that prevents vehicles from passing through, often with a distinct shape or color.
3. **Image Analysis**:
- **Shape and Color**: The coordinates are given in a specific format, which is not typical for a traffic sign. A traffic sign usually has a recognizable shape and color.
- **Barrier Shape**: The coordinates are given in a format that is not typical for a road barrier. Road barriers are usually rectangular or rectangular-like structures.
4. **Contextual Clues**:
- The coordinates are given in a format that is not typical for a traffic sign or road barrier. The coordinates are likely to be from a camera or sensor, which is common in surveillance or mapping applications.
- The coordinates are given in a specific format, which is not typical for a traffic sign or road barrier.
Based on the analysis, the coordinates <c1,CAM_BACK,1159.2,495.0> are most likely a result of a camera or sensor's output, not a traffic sign or road barrier.
Fine-tuned 25kours✓ correct
No.
Fine-tuned 300kours✓ correct
No.
OpenGVLab Mini-DAexternal reference✓ correct
No.
LLaVA-1.6 7Bzero-shotno direct answer
The image you've provided appears to be a collage of different scenes, and it's not clear what the specific context is for the sign or barrier labeled as <c1,CAM_BACK,1159.2,495.0>. However, based on the visible elements, it seems to be a traffic sign or a road barrier, as it is placed on the side of a road and is likely intended to provide information or guidance to drivers.
LLaMA-Adapter v2 7Bzero-shotno direct answer
The image does not provide enough information to confidently determine whether the object is a traffic sign or a road barrier. However, it is clear that the object is located on the side of the road and is part of the street scene.
Q5planning
What actions could the ego vehicle take based on <c1,CAM_BACK,1159.2,495.0>? Why take this action and what's the probability?
Ground truth
The action is to keep going at the same speed because there is no safety issue. The probability of taking this action is high.
Best modelours
The action is to keep going at the same speed. The reason is that there is no safety issue. The probability is high.
Oracleours · upper bound
The action is to keep going at the same speed. The reason is that there is no safety issue. The probability is high.
InternVL2-2B (no fine-tuning)zero-shot
The ego vehicle, indicated by the vehicle icon in the center of the image, is currently positioned at the intersection of a two-lane road. The action it can take is to proceed straight or turn left.
### Chain of Thought (CoT) Analysis:
1. **Current Position**: The ego vehicle is at the intersection of a two-lane road.
2. **Intersection**: The intersection is marked with a pedestrian crossing sign, indicating that vehicles must yield to pedestrians.
3. **Direction**: The ego vehicle is currently facing the intersection, which means it must decide whether to proceed straight or turn left.
4. **Probability**: The probability of proceeding straight or turning left depends on the current traffic conditions and the intentions of other vehicles and pedestrians.
### Possible Actions:
- **Proceed Straight**: If the ego vehicle is in a clear lane and there are no other vehicles or pedestrians in the immediate vicinity, proceeding straight is a reasonable action. This action is likely to be safe and avoids any potential conflicts with pedestrians.
- **Turn Left**: If the ego vehicle is in a lane that is not clear of other vehicles or pedestrians, turning left could be a safer option. This action minimizes the risk of a collision with pedestrians or other vehicles.
### Conclusion:
The ego vehicle should consider the following factors:
- **Pedestrian Crossing**: The intersection has a pedestrian crossing sign, indicating that vehicles must yield to pedestrians.
- **Lane Clearance**: The ego vehicle must ensure it has enough space to proceed straight or turn left without obstructing the pedestrian crossing.
- **Traffic Conditions**: Assessing the current traffic conditions, including the presence of other vehicles and pedestrians, to determine the most appropriate action.
Given these considerations, the ego vehicle should proceed straight if it is clear of any obstructions and has enough space, or turn left if it is in a clear lane and there are no immediate hazards.
Fine-tuned 25kours
The action is to keep going at the same speed. The reason is that there is no safety issue, and the probability is high.
Fine-tuned 300kours
The action is to keep going at the same speed. The reason is that there is no safety issue. The probability is high.
OpenGVLab Mini-DAexternal reference
The action is to keep going at the same speed. The reason is that there is no safety issue. The probability is high.
LLaVA-1.6 7Bzero-shot
The image you've provided appears to be a sequence of images showing a vehicle's perspective as it moves through an urban environment. The sequence includes a "CAM_BACK" label, which suggests that the vehicle is using a rear-facing camera to monitor its surroundings.
Based on the label <c1,CAM_BACK,1159.2,495.0>, the ego vehicle could take the following actions:
1. **Slow down or stop**: The vehicle might need to slow down or stop if it detects an obstacle or a pedestrian in its path. The probability of this action is high if there is a clear obstacle or if the vehicle's sensors detect a pedestrian or another vehicle too close.
2. **Change lanes**: If the vehicle is in a multi-lane road and there is enough space and no oncoming traffic, it could change lanes to avoid an obstacle or to maintain a safe distance from other vehicles. The probability of this action depends on the traffic conditions and the vehicle's lane-changing policy.
3. **Turn**: If the vehicle needs to make a turn, it could do so. The probability of this action depends on the vehicle's intended destination and the road layout.
4. **Maintain speed**: If the vehicle is maintaining a safe speed and there are no immediate hazards, it could continue driving at its current speed. The probability of this action is high if the vehicle's sensors do not detect any immediate danger.
The exact action the vehicle should take would depend on the specific context of the image, which includes the vehicle's speed, the road conditions, the presence of other vehicles or pedestrians, and the vehicle's programming for safe driving.
LLaMA-Adapter v2 7Bzero-shot
Based on the given coordinates, the ego vehicle is likely to take a left turn at the intersection. The vehicle is currently at the intersection, and the coordinates indicate that it is at the left side of the intersection. The probability of the vehicle taking a left turn is high, as it is approaching the intersection from the left side and is already in the correct lane for a left turn. The vehicle would need to follow the traffic rules and signals to ensure a safe and smooth turn.
Q6planning
What actions taken by the ego vehicle can lead to a collision with <c1,CAM_BACK,1159.2,495.0>?
Ground truth
No such action will lead to a collision.
Best modelours
No such action will lead to a collision.
Oracleours · upper bound
No such action will lead to a collision.
InternVL2-2B (no fine-tuning)zero-shot
The ego vehicle is approaching a road with a pedestrian crossing. The pedestrian crossing is located at the intersection of the road, and the pedestrian signal is visible. The pedestrian signal is green, indicating that it is safe for pedestrians to cross the road. The ego vehicle should have stopped at the pedestrian crossing to ensure the safety of the pedestrian. If the ego vehicle were to continue driving, it could potentially collide with the pedestrian crossing due to the lack of stopping at the signal.
Fine-tuned 25kours
Back up.
Fine-tuned 300kours
Back up.
OpenGVLab Mini-DAexternal reference
Back up.
LLaVA-1.6 7Bzero-shot
The image you've provided appears to be a series of snapshots from a vehicle's camera system, showing different views of the road ahead. To avoid a collision with the vehicle labeled as <c1,CAM_BACK,1159.2,495.0>, the ego vehicle should take the following actions:
1. **Maintain a Safe Following Distance**: The ego vehicle should ensure it maintains a safe distance from the vehicle in front, as indicated by the "CAM_BACK" label. This is important to allow enough time to react to any sudden changes in the vehicle's speed or direction.
2. **Stay in the Lane**: The ego vehicle should stay within its lane, as indicated by the "CAM_FRONT_LEFT" label. This is crucial to avoid collisions with other vehicles or obstacles on the road.
3. **Be Aware of Surroundings**: The ego vehicle should be aware of its surroundings, including the presence of other vehicles, pedestrians, and potential obstacles. This is evident from the "CAM_FRONT_RIGHT" label, which shows the right side of the road.
4. **Anticipate Turns**: The ego vehicle should anticipate any turns that the vehicle in front might make, as indicated by the "CAM_FRONT_LEFT" label. This will help the ego vehicle to adjust its speed and position accordingly.
5. **Use Turn Signals**: If the ego vehicle intends to make a turn, it should use its turn signals to communicate its intentions to other road users.
6. **Slow Down When Necessary**: If the ego vehicle detects that the vehicle in front is slowing down or stopping, it should also slow down or prepare to stop to avoid a collision.
7. **React to Traffic Signs**: The ego vehicle should be aware of and react to any traffic signs or signals, such as the ones visible in the "CAM_FRONT_LEFT" and "CAM_FRONT_RIGHT" labels.
8. **Follow Traffic Rules**: The ego vehicle should follow all traffic rules and regulations, including speed limits, right-of-way, and lane changes.
By following these guidelines, the ego vehicle can significantly reduce the risk of a collision with the vehicle labeled as <c1,CAM_BACK,1159.2,4
LLaMA-Adapter v2 7Bzero-shot
The ego vehicle is driving down the street, and it is approaching a red traffic light at an intersection. If the driver of the ego vehicle does not stop or slow down in time, they may collide with the red traffic light. This can lead to accidents, property damage, and potential injuries to the vehicle occupants or other road users. It is crucial for drivers to obey traffic signals and follow traffic rules to ensure the safety of all road users.
Q7planning
In this scenario, what are safe actions to take for the ego vehicle?
Ground truth
Keep going at the same speed, decelerate gradually without braking.
Best modelours
Keep going at the same speed, decelerate gradually without braking.
Oracleours · upper bound
Keep going at the same speed, accelerate, and proceed.
InternVL2-2B (no fine-tuning)zero-shot
Based on the image, the safe actions for the ego vehicle would include maintaining a safe following distance from the vehicle in front, ensuring that the vehicle in front is not overtaking, and being prepared to stop if necessary. The ego vehicle should also be aware of the speed limit and any potential hazards ahead, such as pedestrians or other vehicles, and adjust its speed accordingly. Additionally, the ego vehicle should be vigilant for any changes in traffic patterns or road conditions that may affect its path.
Fine-tuned 25kours
Keep going at the same speed, accelerate, and proceed.
Fine-tuned 300kours
Keep going at the same speed, decelerate gradually without braking.
OpenGVLab Mini-DAexternal reference
Keep going at the same speed, decelerate gradually without braking.
LLaVA-1.6 7Bzero-shot
In the given scenario, the ego vehicle is driving down a road with various types of traffic and pedestrians. Here are some safe actions the ego vehicle should take:
1. **Maintain a Safe Distance**: The ego vehicle should keep a safe distance from other vehicles, pedestrians, and obstacles to avoid collisions.
2. **Follow Traffic Rules**: The vehicle should obey all traffic signals, signs, and road markings. This includes stopping at red lights, yielding to pedestrians, and staying within the designated lanes.
3. **Be Alert for Pedestrians**: The image shows several pedestrians near the road. The ego vehicle should be prepared to stop or slow down if necessary to avoid accidents involving pedestrians.
4. **Anticipate Turns**: The road appears to have multiple turns. The ego vehicle should be prepared to slow down and yield to vehicles making turns.
5. **Use Turn Signals**: If the ego vehicle is making a turn, it should use the appropriate turn signals to communicate its intentions to other road users.
6. **Stay Within Speed Limits**: The vehicle should adhere to the speed limits to ensure safe driving conditions.
7. **Be Prepared for Sudden Changes**: The image shows a variety of road conditions, including different types of vehicles and pedestrians. The ego vehicle should be prepared for sudden changes in traffic flow or unexpected obstacles.
8. **Use Proper Lane Changing Signals**: If the ego vehicle needs to change lanes, it should use the appropriate lane change signals to communicate its intentions to other drivers.
9. **Stay Focused**: The driver should stay focused on the road and avoid distractions such as using a mobile phone or engaging in other activities that could divert attention from the road.
10. **Regularly Check Surroundings**: The driver should regularly check their mirrors and the road ahead to ensure they are aware of their surroundings and any potential hazards.
By following these safety guidelines, the ego vehicle can navigate the road safely and responsibly.
LLaMA-Adapter v2 7Bzero-shot
In the image, there are multiple cars on the road, including a white car and a truck. The white car is driving down the street, and the truck is parked on the side of the road. To ensure safety, the ego vehicle should follow the speed limit, maintain a safe distance from other vehicles, and be prepared to react to any sudden changes in traffic conditions. Additionally, the driver should be attentive to the traffic lights and follow the signals to avoid accidents. It is also essential to be cautious when approaching parked vehicles, as they may suddenly open their doors or move out of the way, requiring the driver to adjust their speed and direction accordingly.
Q8behaviormultiple choice
Predict the behavior of the ego vehicle. Please select the correct answer from the following options: A. The ego vehicle is slightly steering to the left. The ego vehicle is driving with normal speed. B. The ego vehicle is slightly steering to the left. The ego vehicle is driving slowly. C. The ego vehicle is going straight. The ego vehicle is driving very fast. D. The ego vehicle is slightly steering to the right. The ego vehicle is driving with normal speed.
Ground truth
C
Best modelours✓ correct
C
Oracleours · upper bound✓ correct
C
InternVL2-2B (no fine-tuning)zero-shotno direct answer
Based on the image, the ego vehicle is driving straight. The road markings and the absence of any other vehicles or signs of deviation suggest that the ego vehicle is maintaining a straight path.
Fine-tuned 25kours✓ correct
C
Fine-tuned 300koursno direct answer
D. The ego vehicle is slightly steering to the right. The ego vehicle is driving with normal speed.
OpenGVLab Mini-DAexternal reference✓ correct
C
LLaVA-1.6 7Bzero-shotno direct answer
A. The ego vehicle is slightly steering to the left. The ego vehicle is driving with normal speed.
LLaMA-Adapter v2 7Bzero-shotno direct answer
The ego vehicle is driving with normal speed, and it is slightly steering to the left.