id stringlengths 11 17 | source stringclasses 2
values | source_id stringlengths 4 6 | category stringclasses 88
values | collision bool 2
classes | ground_truth stringlengths 36 210 |
|---|---|---|---|---|---|
MMAU_000848 | MM-AU | 000848 | car scratches truck | true | A brown SUV tries to change lanes near the toll booth and ends up hitting the back corner of a tanker truck. |
MMAU_000855 | MM-AU | 000855 | motorbike falls down | false | A motorcyclist suddenly loses control and the bike topples over onto the highway. |
MMAU_000875 | MM-AU | 000875 | truck hits car | true | A car loses control while turning and collides with a loaded truck. |
MMAU_000880 | MM-AU | 000880 | ego-car is out of control | false | The dash-cam vehicle has lost control and is spinning/skidding across the roadway. |
MMAU_000909 | MM-AU | 000909 | motorbike scratches car | true | A motorbike collides with the suddenly opened door of a white car, scraping it and throwing the rider to the ground. |
MMAU_000929 | MM-AU | 000929 | car hits trees | true | A white car veers off the road and crashes into the line of roadside trees. |
MMAU_000930 | MM-AU | 000930 | motorbike scratches car | true | A car tries to change lanes and hits a motorbike, causing the motorbike to crash into the dash-cam car. |
MMAU_000931 | MM-AU | 000931 | car hits telegraphpoles | true | A white car suddenly mounts the traffic island and smashes into two utility poles, snapping them and scattering debris across the intersection. |
MMAU_000975 | MM-AU | 000975 | motorbike falls down | false | A motorcycle loses control and skids onto its side, sliding dangerously across the lane in front of the dash-cam vehicle. |
MMAU_001016 | MM-AU | 001016 | truck hits motorbike | true | A motorbike tries to cross the road and collides with a cargo truck, knocking the rider and bike down in the middle of the road. |
MMAU_001051 | MM-AU | 001051 | ego-car is overtaken by a truck | false | The dash-cam car tries to turn right and almost collides with a blue truck driving in the right lane. |
MMAU_001063 | MM-AU | 001063 | motorbike hits motorbike | true | Two motorcycles collide in the middle of the intersection, sending both riders crashing to the roadway. |
MMAU_001071 | MM-AU | 001071 | motorbike hits motorbike | true | A motorbike crashes into another motorbike in the middle of the intersection. |
MMAU_001106 | MM-AU | 001106 | car hits small roadblocks road pothholes | true | A white car has collided with and knocked over the small concrete roadblocks sitting in the middle of the roadway. |
MMAU_001112 | MM-AU | 001112 | car hits large roadblocks | true | A car suddenly loses control on the snowy highway, veers off-lane, and crashes into the large snow-covered roadside barriers. |
MMAU_001128 | MM-AU | 001128 | truck hits trees | true | A red truck veers into the planted median and smashes into roadside trees, sending large branches crashing onto the roadway. |
MMAU_001143 | MM-AU | 001143 | ego-car hits a motorbike | true | The dash-cam shows the car colliding head-on with a motorcyclist while turning left. |
MMAU_001167 | MM-AU | 001167 | ego-car hits a car | true | The dash-cam (ego) vehicle nearly collides head-on with an oncoming white car on the narrow road. |
MMAU_001184 | MM-AU | 001184 | motorbike scratches car | true | A motorbike tries to turn left and crashes into the side of the car. |
MMAU_001201 | MM-AU | 001201 | ego-car is overtaken by a car | false | A green taxi abruptly swerves past and cuts directly in front of the dash-cam car while overtaking, creating a near-collision situation. |
MMAU_001209 | MM-AU | 001209 | truck is out of control | false | A small green dump truck suddenly skids across the wet road and flips onto its side, blocking the lane. |
MMAU_001221 | MM-AU | 001221 | car hits a curb | true | An SUV loses control on the icy road and suddenly skids sideways across the path of the dash-cam car. |
MMAU_001257 | MM-AU | 001257 | motorbike falls down | false | A motorcyclist ahead loses control and the bike tips over onto its side in the middle of the intersection. |
MMAU_001263 | MM-AU | 001263 | ego-car hits a crossing car | true | The dash-cam (ego) vehicle drives through the intersection and collides with a car turning in front of it. |
MMAU_001282 | MM-AU | 001282 | car hits car | true | A car driving in the opposite direction swerves to avoid another that suddenly emerges from the side street, resulting in a collision with the dashcam-car. |
MMAU_001299 | MM-AU | 001299 | ego-car is overtaken by a truck | false | A large dump truck abruptly cuts in front of the dash-cam car at the stop line, overtaking it through the intersection and resulting in a collision. |
MMAU_001308 | MM-AU | 001308 | motorbike hits motorbike | true | A fast-moving scooter crashes into another scooter in the intersection, knocking both riders to the ground. |
MMAU_001404 | MM-AU | 001404 | motorbike scratches car | true | A scooter slides into the side of a sedan, toppling over and throwing the riders onto the street. |
MMAU_001433 | MM-AU | 001433 | car hits other road facilities | true | An oncoming white car suddenly veers left, scrapes against a scooter and crashes into the roadside construction signs/barriers, creating a roadway hazard. |
MMAU_001448 | MM-AU | 001448 | ego-car is overtaken by a car | false | A black sedan suddenly squeezes past and overtakes the ego-car on the right in a very narrow, single-lane section, creating a dangerous close-quarters situation. |
MMAU_001461 | MM-AU | 001461 | car hits small roadblocks road pothholes | true | A car suddenly veers left and crashes through a row of small roadside barrier posts on the highway. |
MMAU_001465 | MM-AU | 001465 | car hits car | true | A white van suddenly cuts across the road and crashes into a dark SUV in the adjacent lane. |
MMAU_001466 | MM-AU | 001466 | ego-car hits a crossing car | true | The dash-cam shows the ego vehicle crashing into a car that suddenly crosses in front of it from the left. |
MMAU_001516 | MM-AU | 001516 | ego-car is overtaken by a truck | false | The dash-cam car suddenly swerves left and sideswipes a truck, smashing its mirror and showering the windscreen with debris. |
MMAU_001528 | MM-AU | 001528 | car hits car | true | The unusual or dangerous event happening in this scene is: a car drives into the side of a tram as it crosses the intersection. |
MMAU_001550 | MM-AU | 001550 | truck has failure of components | false | A box truck’s rear axle/wheels break off and roll loose down the highway, leaving the crippled truck skidding in the lane. |
MMAU_001578 | MM-AU | 001578 | truck hits large roadblocks | true | A large truck suddenly veers right and smashes into the big roadside barriers, sending up a cloud of dust and debris. |
MMAU_001604 | MM-AU | 001604 | ego-car hits a crossing truck | true | The dash-cam car crashes into a truck that is crossing in front of it. |
MMAU_001624 | MM-AU | 001624 | car hits trees | true | A car loses control on the snowy road and crashes into the trees at the side. |
MMAU_001625 | MM-AU | 001625 | car is out of control | false | A car has lost control and is sliding sideways across the snowy roadway. |
MMAU_001627 | MM-AU | 001627 | ego-car is overtaken by a car | false | A white sedan abruptly cuts across and overtakes directly in front of the dash-cam car, resulting in a collision. |
MMAU_001632 | MM-AU | 001632 | ego-car is overtaken by a car | false | A car cuts in and quickly overtakes the dash-cam vehicle from the right, seemingly bumping the dashcam vehicle. |
MMAU_001652 | MM-AU | 001652 | truck has failure of components | false | A truck in the right-hand lane has lost one of its wheels, which is now rolling loose down the highway and creating a dangerous obstacle for the surrounding traffic. |
MMAU_001676 | MM-AU | 001676 | ego-car hits a curb | true | The car ahead suddenly brakes, forcing the dash-cam car to veer onto the snowy roadside curb. |
MMAU_001682 | MM-AU | 001682 | ego-car hits a crossing car | true | The dash-cam car crashes into a silver coupe that suddenly cuts across its lane. |
MMAU_001686 | MM-AU | 001686 | truck is out of control | false | A truck ahead suddenly swerves across lanes, loses control, and rolls over onto its side off the roadway. |
MMAU_001728 | MM-AU | 001728 | truck hits telegraphpoles | true | The red truck swerved off the road to avoid the blue truck, which was driving on the wrong side. |
MMAU_001752 | MM-AU | 001752 | ego-car is overtaken by a truck | false | A large truck dangerously overtakes the dash-cam car on a dark, narrow road while an oncoming vehicle approaches head-on. |
MMAU_001757 | MM-AU | 001757 | ego-car hits a curb | true | The dash-cam vehicle loses control on the icy road and slams into the roadside curb, resulting in a sudden crash. |
MMAU_001760 | MM-AU | 001760 | car is out of control | false | A silver SUV loses control on the icy overpass, spinning sideways across the lanes and sliding directly toward the dash-cam car. |
MMAU_001788 | MM-AU | 001788 | ego-car is out of control | false | The dash-cam (ego) car suddenly loses traction on the icy road, skids out of control, and crashes. |
MMAU_001793 | MM-AU | 001793 | truck hits car | true | A large truck ahead suddenly brakes, causing the dash-cam vehicle to crash into the truck. |
MMAU_001803 | MM-AU | 001803 | ego-car hits trees | true | The dash-cam vehicle suddenly leaves the roadway and crashes into trees lining the side of the road. |
MMAU_001807 | MM-AU | 001807 | pedestrian falls down | false | A pedestrian slips on the icy road while crossing and suddenly falls down in front of the approaching vehicle. |
MMAU_001822 | MM-AU | 001822 | ego-car is out of control | false | The dash-cam vehicle is swerving wildly, indicating it has lost control on the roadway. |
MMAU_001931 | MM-AU | 001931 | ego-car hits trees | true | The dash-cam car suddenly leaves the roadway and crashes into roadside trees, shattering its windshield. |
MMAU_001942 | MM-AU | 001942 | car is out of control | false | A vehicle ahead suddenly veers off the road, flips over, and skids to a stop upside-down across the highway. |
MMAU_001963 | MM-AU | 001963 | ego-car is overtaken by a car | false | A black Peugeot suddenly speeds up from the right-hand lane to overtake the dash-cam car, causing a collision. |
MMAU_001977 | MM-AU | 001977 | ego-car is out of control | false | The dash-cam (ego) car suddenly loses control, veers off the roadway, and crashes into a roadside fence. |
MMAU_001988 | MM-AU | 001988 | ego-car is out of control | false | The dash-cam vehicle suddenly swerves out of its lane toward the shoulder, indicating the driver has lost control. |
MMAU_001991 | MM-AU | 001991 | car is out of control | false | A white sedan suddenly skids on the icy roadway and spins out of control toward the roadside. |
MMAU_002000 | MM-AU | 002000 | car scratches truck | true | A small car skids while overtaking on the icy road and scrapes along the side of the truck. |
MMAU_002002 | MM-AU | 002002 | car hits a curb | true | An oncoming car skids on the icy highway, slams into the roadside curb/guardrail, then collides with another car, throwing up a cloud of snow and debris. |
MMAU_002008 | MM-AU | 002008 | ego-car is out of control | false | The dash-cam car suddenly loses control on the icy road, skids into the oncoming lane in front of a bus, then slides off onto the snowy shoulder. |
MMAU_002009 | MM-AU | 002009 | car is out of control | false | A car on the icy highway suddenly spins sideways across the lanes, clearly out of control and creating a serious hazard. |
MMAU_002031 | MM-AU | 002031 | car scratches truck | true | An orange dump truck suddenly swerves into the right-hand lane, sideswipes, and scrapes with a white sedan. |
MMAU_002034 | MM-AU | 002034 | truck has failure of components | false | A package suddenly falls off an oncoming truck, bounces along the roadway toward the dash-cam vehicle. |
MMAU_002035 | MM-AU | 002035 | truck hits cyclist | true | A large tanker truck sideswipes a bicyclist, knocking the rider off the bike and onto the roadway. |
MMAU_002046 | MM-AU | 002046 | motorbike falls down | false | A black SUV suddenly swerves to the left, causing an oncoming motorcycle to be unable to brake in time and to collide with the SUV's side. |
MMAU_002086 | MM-AU | 002086 | ego-car is out of control | false | The dash-cam vehicle suddenly veers across the lanes and scrapes the guardrail, indicating it has gone out of control. |
MMAU_002136 | MM-AU | 002136 | truck has failure of components | false | A logging truck has toppled and scattered its load of timber across both lanes, blocking the highway and creating a major hazard. |
MMAU_002137 | MM-AU | 002137 | truck has failure of components | false | A truck ahead suddenly sheds a rear wheel, which bounces off down the road while the crippled truck swerves dangerously. |
MMAU_002145 | MM-AU | 002145 | car is out of control | false | A dark sedan suddenly skids sideways, slides off the roadway into the grass and flips over beside a utility pole. |
MMAU_002150 | MM-AU | 002150 | ego-car is overtaken by a car | false | A car abruptly overtakes the dash-cam vehicle, cuts in dangerously close, stops in the lane, and the driver then steps out into the middle of moving traffic. |
MMAU_002167 | MM-AU | 002167 | truck hits car | true | An oncoming car suddenly drifts across the center line and crashes head-on into a truck. |
MMAU_002169 | MM-AU | 002169 | truck hits truck | true | A truck collides with another truck. |
MMAU_002176 | MM-AU | 002176 | motorbike hits motorbike | true | A scooter rear-ends another scooter while overtaking, sending both riders sliding onto the roadway in the intersection. |
MMAU_002178 | MM-AU | 002178 | car hits cyclist | true | A turning car strikes a cyclist in the intersection, knocking the rider and bicycle onto the roadway. |
MMAU_002179 | MM-AU | 002179 | car hits cyclist | true | A car making a left turn strikes a cyclist crossing the road, throwing the rider onto the pavement. |
MMAU_002180 | MM-AU | 002180 | cyclist falls down | false | A cyclist loses control while riding across the crosswalk and crashes onto the roadside. |
MMAU_002194 | MM-AU | 002194 | ego-car is out of control | false | The dash-cam car loses control on the icy, snow-covered road and skids off into the snowy embankment. |
MMAU_002195 | MM-AU | 002195 | car is out of control | false | A small hatchback suddenly skids sideways across the icy roadway, spinning out of control in front of the dash-cam vehicle. |
MMAU_002209 | MM-AU | 002209 | cyclist falls down | false | A cyclist has lost control and fallen onto the roadway in front of approaching traffic. |
MMAU_002210 | MM-AU | 002210 | motorbike hits motorbike | true | A scooter crashes into another motorbike in the middle of the intersection, knocking one bike and its passengers to the ground. |
MMAU_002211 | MM-AU | 002211 | motorbike falls down | false | A motorcycle loses control and tips over, sending its riders tumbling onto the road directly in front of the dash-cam vehicle. |
MMAU_002253 | MM-AU | 002253 | car is out of control | false | A car skids on the icy road, spins sideways, and slides off onto the snowy sidewalk. |
MMAU_002280 | MM-AU | 002280 | car hits trees | true | A vehicle swerves off the roadway and crashes into the roadside trees. |
MMAU_002299 | MM-AU | 002299 | motorbike hits motorbike | true | A blue scooter smashes head-on into another scooter in the intersection, throwing both riders onto the road. |
MMAU_002304 | MM-AU | 002304 | motorbike scratches car | true | A motorbike veers across the road and scrapes against a passing car, creating a dangerous situation. |
MMAU_002305 | MM-AU | 002305 | motorbike hits other road facilities | true | A motorbike loses control and crashes into the roadside fence/guardrail. |
MMAU_002306 | MM-AU | 002306 | truck has failure of components | false | A truck ahead suddenly sheds a rear wheel, scattering debris and forcing it to swerve toward the shoulder. |
MMAU_002319 | MM-AU | 002319 | car hits telegraphpoles | true | A van veers off the roadway and smashes into the line of telegraph-poles along the right-hand side. |
MMAU_002320 | MM-AU | 002320 | car has failure of components | false | A car’s wheel suddenly detaches and rolls/bounces down the road, creating a dangerous hazard for the surrounding traffic. |
MMAU_002338 | MM-AU | 002338 | ego-car is out of control | false | The dash-cam shows the ego-vehicle sliding off the muddy road and losing control. |
MMAU_002342 | MM-AU | 002342 | ego-car is overtaken by a truck | false | The dash-cam car swerves left to avoid the van ahead, resulting in a collision with a large truck coming from the left-hand lane. |
MMAU_002351 | MM-AU | 002351 | ego-car hits small roadblocks road pothholes | true | The dash-cam car drives over small road obstacles on the roadway, causing a jolt. |
MMAU_002355 | MM-AU | 002355 | car scratches truck | true | A white sedan veers across the striped buffer zone and scrapes against the purple semi-truck as they merge toward the toll gate. |
MMAU_002375 | MM-AU | 002375 | ego-car hits trees | true | The dash-cam vehicle veers off the roadway and crashes into the trees lining the side. |
MMAU_002398 | MM-AU | 002398 | motorbike scratches car | true | A motorcycle clips the side of a turning white car in the intersection and skids out, leaving the rider and bike sprawled in the roadway. |
MMAU_002429 | MM-AU | 002429 | truck hits a curb | true | A flat-bed truck loses control on the icy road, jack-knifes and its trailer smashes into the snowy roadside curb/ditch. |
DashBench: Evaluating Video LLMs' Understanding of Corner Cases in Autonomous Driving
Gabriel Manalu, Minh Son Hoang, Hyeontaek Hwang, Jaesung Lee, Hyojin Choi, Tran Tuan Phong, Seon Kwon Kim, Saad Wazir, Daeyoung Kim
School of Computing, KAIST, Daejeon, Korea
BMVC 2026
Abstract
Natural language explanations promised by modern Video Large Language Models (Video LLMs) offer autonomous driving systems a path toward detailed scene description and interpretable decision making on the road. While recent benchmarks have begun assessing these models on driving scenes, they mainly target routine traffic conditions and general comprehension tasks based on existing driving datasets and therefore fail to evaluate model performance on the rarest and most safety-critical driving events, also known as "corner cases". To provide a comprehensive evaluation of Video LLMs on precisely these scenarios, we introduce DashBench, a benchmark testing 25 Video LLMs on 1,709 carefully curated driving videos spanning not only collision-related corner cases but also rare non-collision ones. Through extensive prompts and open-ended question answering, we expose a critical capability gap: all evaluated models perform substantially below the threshold required for safety-critical driving applications, frequently misjudging or missing dangerous events on the road ahead and hallucinating non-existent agents or actions in their explanations.
Dataset Overview
| Clips | 1,709 ego-view dashcam clips |
| Categories | 88 (46 collision, 42 non-collision) |
| Collision / non-collision | 890 / 819 clips |
| Sources | MM-AU (1,395 clips), RoadSocial (314 clips) |
| Frames per clip | 16, uniformly sampled (full) or within the corner-case window (ccw, 823 clips) |
| Prompts per clip | 5 open-ended prompts (Q1–Q5) |
Each clip was manually verified to contain a visually identifiable, self-contained corner case, annotated with keyframes (abnormal start, most observable, abnormal end), and given a one-sentence ground-truth description drafted with OpenAI o3 and refined to consensus by three annotators.
Configurations and splits
| Config | Split | Rows | Description |
|---|---|---|---|
default |
full |
1,709 | All clips. Frames: 16 sampled uniformly across the full clip. |
default |
ccw |
823 | Corner-case-window subset: the same kind of clip, but the 16 frames are sampled inside the annotated corner-case window only (the temporal-localization experiment in the paper). |
prompts |
train |
5 | The Q1–Q5 prompts, the system prompt, and the judge prompt. |
One row is one clip. Every clip is evaluated with all five prompts, so a clip is stored once instead of once per clip–prompt pair (the benchmark has 1,709 × 5 = 8,545 clip–prompt pairs).
The viewer shows the text fields. The 16 frames of each clip (original JPEGs, no re-encoding) are stored separately in frames/*.parquet, with the frame numbers and the annotated keyframes (abnormal start, most observable, abnormal end). Join them to the text rows by id.
frames/full-*.parquet 1,709 clips frames/ccw-*.parquet 823 clips
columns: id, frames (16 images), frame_numbers, abnormal_start_frame, accident_frame, abnormal_end_frame
Fields
| Field | Type | Description |
|---|---|---|
id |
string | Clip id, e.g. MMAU_000848, RoadSocial_RS_1; also the key in frames/*.parquet |
source |
string | MM-AU or RoadSocial |
source_id |
string | Id in the source dataset |
category |
string | One of the corner-case categories |
collision |
bool | Collision vs. non-collision corner case |
ground_truth |
string | Human-verified description of the corner case |
Usage
from datasets import load_dataset
repo = "gabrieldavin23/cornercase-dashbench"
meta = load_dataset(repo, split="full") # or split="ccw"
frames = load_dataset("parquet", data_files={"full": f"hf://datasets/{repo}/frames/full-*.parquet"}, split="full")
prompts = load_dataset(repo, "prompts", split="train")
clip = meta[0]
print(clip["category"], "|", clip["ground_truth"])
imgs = frames[0]["frames"] # 16 PIL images, temporal order (row i of `frames` matches row i of `meta`)
for p in prompts:
question = p["prompt"] # ask your Video LLM this with the 16 frames
Prompts
| Prompt | |
|---|---|
| Q1 | What is unusual in this sequence of frames? |
| Q2 | What is dangerous in this sequence of frames? |
| Q3 | Describe this sequence of frames in detail. |
| Q4 | JSON answer with free-text reasoning (Q4R) and a boolean unusual_or_dangerous_event_detected (Q4TF). |
| Q5 | Defines a corner case with examples, then asks the model to identify and describe one. |
Evaluation
Responses to Q1, Q2, Q3, Q4R and Q5 are scored by DeepSeekV3-as-a-judge, which answers Yes/No to whether the response captures the same corner case as the ground truth (cause, agents responsible, and timing). Accuracy is the share of Yes verdicts. DeepSeekV3 agrees with human judgments at Cohen's κ = 0.96. Q4TF is reported separately as binary detection accuracy. The judge prompt is in the prompts config.
License and source data
DashBench is built from clips of MM-AU and RoadSocial. Use of the frames is subject to the terms of those datasets. The annotations, ground-truth descriptions and prompts are released under CC BY-NC 4.0 for non-commercial research use.
Citation
@misc{manalu2026evaluatingvideollmsunderstanding,
title={Evaluating Video LLMs' Understanding of Corner Cases in Autonomous Driving},
author={Gabriel Manalu and Minh Son Hoang and Hyeontaek Hwang and Jaesung Lee and Hyojin Choi and Tran Tuan Phong and Seon Kwon Kim and Saad Wazir and Daeyoung Kim},
year={2026},
eprint={2611.16767},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2611.16767},
}
- Downloads last month
- 39