---
language:
- en
license: gpl-3.0
tags:
- vision
- image-segmentation
- instance-segmentation
- object-detection
- optical-flow
- depth
- synthetic
- sim-to-real
annotations_creators:
- machine-generated
pretty_name: SMVB Dataset
size_categories:
- 1K<n<10K
task_categories:
- object-detection
- image-segmentation
- depth-estimation
- video-classification
- other
task_ids:
- instance-segmentation
- semantic-segmentation
---

# Synthetic Multimodal Video Benchmark (SMVB)

A dataset consisting of synthetic images from distinct synthetic scenes, annotated with object/instance/semantic segmentation masks, depth data, surface normal information and optical for testing and benchmarking model performance for multi-task/multi-objective learning.

### Supported Tasks and Leaderboards

The dataset supports tasks such as semantic segmentation, instance segmentation, object detection, image classification, depth, surface normal, and optical flow estimation, and video object segmentation.

## Dataset Structure

### Data Instances


### Data Fields

### Data Splits


## Dataset Creation

### Curation Rationale

### Source Data

### Citation Information

```bibtex
@INPROCEEDINGS{karoly2024synthetic,
  author={Károly, Artúr I. and Nádas, Imre and Galambos, Péter},
  booktitle={2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics (SAMI)}, 
  title={Synthetic Multimodal Video Benchmark (SMVB): Utilizing Blender for rich dataset generation}, 
  year={2024},
  volume={},
  number={},
  pages={},
  doi={}}
```