Fast Panoptic Segmentation Network

Geus, Daan de; Meletis, Panagiotis; Dubbelman, Gijs

doi:10.48550/arxiv.1910.03892

Cited by 3 publications

(5 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Publications. [11], FPSNet [12], EPSNet [13] і VPSNet [14]. Наприклад, семантичний розрив між (X 0,0 ,X 1,3 ) нівелюється за допомогою Також в статті запропоноване використання глибокого контрольованого навчання (Deep Supervision [24]), що дозволяє моделі працювати в двох режимах:…”

Section: Practical Value Of Obtained Resultsunclassified

Розпізнавання (Сегментація) Об’єктів На Супутникових Та Аерофотознімках

Снагощенко¹

2022

ВОТТП

View full text Add to dashboard Cite

В роботі на основі проведеного аналізу існуючих методів сегментації зображень та архутектур нейронної мережі, провівши низку експериментів використовуючи Набір даних Inria Aerial Image Labeling створений для основної задачі дистанційного зондування, а саме автоматичного попіксельного маркування аерофотознімків визначено, що саме експерименту U-Net++/EfficientNetB2/BCE+Dice Loss є найбільш вдалим для вирішення поставленої задачі розпізнавання (сегментація) об’єктів на супутникових та аерофотознімках

show abstract

Section: Practical Value Of Obtained Resultsunclassified

Розпізнавання (Сегментація) Об’єктів На Супутникових Та Аерофотознімках

Снагощенко¹

2022

ВОТТП

View full text Add to dashboard Cite

show abstract

“…BB PQ PQ T h PQ St IoU FPSNet [9] 55.1 48.3 60.1 -TASCNet [19] 55.9 50.5 59.8 -AUNet [21] 56.4 52.7 59.0 73.6 P. FPN [16] 57 In Table 4 we benchmark the quantitative performance on the Microsoft COCO dataset while qualitative results are shown in figure 6. Similar to the methodology used for Cityscapes we report results with same backbone and with same pre-training.…”

Section: Methodsmentioning

confidence: 99%

“…Despite the recent introduction of panoptic segmentation there have already been multiple works attempting to address this [9,19,21,41]. This is in part due to its importance to the wider community, success in individual subtasks of instance and semantic segmentation and publicly available datasets to benchmark different methods.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Towards Bounding-Box Free Panoptic Segmentation

Bonde¹,

Alcantarilla²,

Leutenegger³

2020

Preprint

View full text Add to dashboard Cite

In this work we introduce a new bounding-box free network (BBFNet) for panoptic segmentation. Panoptic segmentation is an ideal problem for a bounding-box free approach as it already requires per-pixel semantic class labels. We use this observation to exploit class boundaries from an off-the-shelf semantic segmentation network and refine them to predict instance labels. Towards this goal BBFNet predicts coarse watershed levels and use it to detect large instance candidates where boundaries are well defined. For smaller instances, whose boundaries are less reliable, BBFNet also predicts instance centers by means of Hough voting followed by mean-shift to reliably detect small objects. A novel triplet loss network helps merging fragmented instances while refining boundary pixels. Our approach is distinct from previous works in panoptic segmentation that rely on a combination of a semantic segmentation network with a computationally costly instance segmentation network based on bounding boxes, such as Mask R-CNN, to guide the prediction of instance labels using a Mixture-of-Expert (MoE) approach. We benchmark our non-MoE method on Cityscapes and Microsoft COCO datasets and show competitive performance with other MoE based approaches while outperfroming exisiting non-proposal based approaches. We achieve this while been computationally more efficient in terms of number of parameters and FLOPs. Video results are provided here https://blog.slamcore.com/ reducing-the-cost-of-understanding

show abstract

“…We use the Adam optimizer[22] with learning rate 10 −5 , polynomial learning rate decay, and test-time flip. We use ResNet50 as a backbone for DeepLabV3+ (pretrained on ImageNet) with an embedding dimension of12 on Cityscapes, and 128 on COCO and Vistas. During training, we use crop size 1024 × 2048 on Cityscapes, and 512 × 512 on COCO, and crop around thing classes.…”

mentioning

confidence: 99%

Hierarchical Lovász Embeddings for Proposal-free Panoptic Segmentation

Kerola¹,

Li²,

Kanehira³

et al. 2021

Preprint

View full text Add to dashboard Cite

Panoptic segmentation brings together two separate tasks: instance and semantic segmentation. Although they are related, unifying them faces an apparent paradox: how to learn simultaneously instance-specific and category-specific (i.e. instance-agnostic) representations jointly. Hence, stateof-the-art panoptic segmentation methods use complex models with a distinct stream for each task. In contrast, we propose Hierarchical Lovász Embeddings, per pixel feature vectors that simultaneously encode instance-and categorylevel discriminative information. We use a hierarchical Lovász hinge loss to learn a low-dimensional embedding space structured into a unified semantic and instance hierarchy without requiring separate network branches or object proposals. Besides modeling instances precisely in a proposal-free manner, our Hierarchical Lovász Embeddings generalize to categories by using a simple Nearest-Class-Mean classifier, including for non-instance "stuff" classes where instance segmentation methods are not applicable. Our simple model achieves state-of-the-art results compared to existing proposal-free panoptic segmentation methods on Cityscapes, COCO, and Mapillary Vistas. Furthermore, our model demonstrates temporal stability between video frames.

show abstract

Fast Panoptic Segmentation Network

Cited by 3 publications

References 0 publications

Розпізнавання (Сегментація) Об’єктів На Супутникових Та Аерофотознімках

Розпізнавання (Сегментація) Об’єктів На Супутникових Та Аерофотознімках

Towards Bounding-Box Free Panoptic Segmentation

Hierarchical Lovász Embeddings for Proposal-free Panoptic Segmentation

Contact Info

Product

Resources

About