Benchmarking Deep Learning Models for Instance Segmentation

Jung, Sunguk; Heo, Hyeonbeom; Park, Sangheon; Lee, Kyungjae

doi:10.3390/app12178856

Cited by 17 publications

(5 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent scholarly focus has crystallized around single-stage real-time instance segmentation [21], presenting models that parallel the performance of their weightier non-real-time counterparts. For instance, the YOLACT++ model orchestrates real-time execution by segmenting the task into two discrete sub-tasks and subsequently amalgamating their outputs through a linear operation.…”

Section: Related Workmentioning

confidence: 99%

TrashInst: real-time floating litter instance segmentation model

Bessaad,

Qilian,

Jiankang

et al. 2024

Eng. Res. Express

View full text Add to dashboard Cite

We propose TrashInst, a novel fully convolutional single-stage anchor-free real-time instance segmentation model tailored for detecting floating litter in waterways. TrashInst boasts a streamlined Encoder Decoder architecture and an efficient channel attention mask-head, ensuring the preservation of essential feature maps while minimizing redundant computations. Even in scenarios with limited feature maps, our approach, augmented by sparse instance activation maps, excels in detection and segmentation tasks. Our encoder effectively processes ResNet50-derived features across multiple scales, creating comprehensive abstractions via the nested feature module. Subsequently, the Vortex Instance Activation Maps module (VIAM) aggregates these multi-scale nested features and merges them with mask head outputs, ultimately producing precise instance masks. Furthermore, we employ the focal Tversky objective function to balance the dataset. In terms of performance, our approach outpaces state-of-the-art real-time instance segmentation models, achieving a remarkable 34% accuracy boost, all while maintaining swift execution at 44 frames per second (FPS) on a single NVIDIA GTX 3090 GPU. Rigorous validation underscores the delicate interplay between-speed and precision, highlighting the model's remarkable distinction in handling objects of varying sizes in real-time settings. Significantly, TrashInst attains superior average precision (AP) results in detecting large to medium-sized objects, outperforming existing models by a notable 422% in the large category and an impressive 1.1~56% in the medium category within our dataset."

show abstract

Section: Related Workmentioning

confidence: 99%

TrashInst: real-time floating litter instance segmentation model

Bessaad,

Qilian,

Jiankang

et al. 2024

Eng. Res. Express

View full text Add to dashboard Cite

show abstract

“…For this task to be used in real-time applications, accuracy and inference time are crucial. [19]. Thus, semantic segmentation differs from instance segmentation in that it treats many objects belonging to the same category as a single entity.…”

Section: Segmentationmentioning

confidence: 99%

“…There are a lot of studies involving deep learning techniques and explaining them in details [16]- [18]. All instances of a class are found using instance segmentation, with the added ability to distinguish between distinct instances of any segment class [19]. Segmentation method entered in many medical and biological specialties [20], [21].Tumor brain segmentation interesting task to many author [22], [23].…”

Section: Introductionmentioning

confidence: 99%

MRI Semantic Segmentation based on Optimize V-net with 2D Attention

Al-azzwi,

Nazarov

2023

Adv. sci. technol. eng. syst. j.

View full text Add to dashboard Cite

Over the past ten years, deep learning models have considerably advanced research in artificial intelligence, particularly in the segmentation of medical images. One of the key benefits of medical picture segmentation is that it allows for a more accurate analysis of anatomical data by separating only pertinent areas. Numerous studies revealed that these models could make accurate predictions and provide results that were on par with those of doctors. In this study, we investigate different methods of deep learning with medical image segmentation, like the V-net and U-net models. Improve the V-net model by adding attention in 2D with a decoder to get high performance through the training model. Using tumors of severe forms, size, and location, we downloaded the BRAST 2018 data set from Kaggle and manually segmented structural T1, T1ce, T2, and Flair MRI images. To enhance segmentation performance, we also investigated several benchmarking and preprocessing procedures. It's significant to note that our model was applied on Colab-Google for 35 epochs with a batch size of 8. In conclusion, we offer a memory-effective and effective tumor segmentation approach to aid in the precise diagnosis of oncological brain diseases. We have tested residual connections, decoder attention, and deep supervision loss in a comprehensive ablation study. Also, we looked for the U-Net encoder and decoder depth, convolutional channel count, and post-processing approach.

show abstract

“…Instance segmentation, which combines the tasks of object detection and segmentation [20], has evolved into a technique in its own right. It is characterized by its ability to identify and separate individual objects of the same class, which enables detailed examination at the pixel level [19,21].…”

Section: Related Workmentioning

confidence: 99%

Distance Estimation Approach for Maritime Traffic Surveillance Using Instance Segmentation

Petković,

Vujović

2023

JMSE

View full text Add to dashboard Cite

Maritime traffic monitoring systems are particularly important in Mediterranean ports, as they provide more comprehensive data collection compared to traditional systems such as the Automatic Identification System (AIS), which is not mandatory for all vessels. This paper improves the existing real-time maritime traffic monitoring systems by introducing a distance estimation algorithm for monocular cameras, which aims to provide high quality maritime traffic metadata collection for traffic density analysis. Two distance estimation methods based on a pinhole camera model are presented: the Vessel-Focused Distance Estimation (VFDE) and the novel Vessel Object-Focused Distance Estimation (VOFDE). While VFDE uses the predefined height of a vessel for distance estimation, VOFDE uses standardized dimensions of objects on the vessel, detected with a Convolutional Neural Network (CNN) for instance segmentation to enhance estimation accuracy. Our evaluation covers distances up to 414 m, which is significantly beyond the scope of previous studies. When compared to the distances measured with a precise instrument, VOFDE achieves a Percentage Deviation Index (PDI) of 1.34% to 9.45%. This advance holds significant potential for improving maritime surveillance with monocular cameras and is also applicable in other areas, such as low-cost maritime vehicles equipped with single cameras.

show abstract

Benchmarking Deep Learning Models for Instance Segmentation

Cited by 17 publications

References 23 publications

TrashInst: real-time floating litter instance segmentation model

TrashInst: real-time floating litter instance segmentation model

MRI Semantic Segmentation based on Optimize V-net with 2D Attention

Distance Estimation Approach for Maritime Traffic Surveillance Using Instance Segmentation

Contact Info

Product

Resources

About