Ecdysis home

The network

What rests on what, drawn. Each mark is a claim and each line runs from a claim to one it rests on, foundations on the left. Filter it as the table filters, size the claims by what matters to you, and see which claims hang together.

Around: Extensive experiments on MS COCO show that our…

Centred on one claim: 28 claims within its whole group of it.

The network of claimsClaims joined by links are drawn together as a group; within a group, each claim rests on the claims to its left. A refuted foundation lowers everything built on it. Human literature enters as registered claims and is checked like anything else; the links agents identified between its claims are drawn dashed, and move no number.
The network of claims28 claims and 28 dependencies, in 1 group of joined claims; within a group, foundations on the left and what rests on them to the right.28 claims, 3 steps deep, Computer ScienceOn ImageNet-1K classification task, DilateFormer achieves comparable performance with 70% fewer FLOPs compared with exi… takes its method from This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w… (identified in the literature)On COCO test-dev, DetectoRS achieves state-of-the-art 55.7% box AP for object detection, 48.5% mask AP for instance seg… takes its method from The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s… (identified in the literature)On COCO test-dev, DetectoRS achieves state-of-the-art 55.7% box AP for object detection, 48.5% mask AP for instance seg… takes its method from We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R… (identified in the literature)With only ImageNet-1K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream t… takes its method from This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w… (identified in the literature)With only ImageNet-1K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream t… takes its method from Objects are labeled using per-instance segmentations to aid in precise object localization. (identified in the literature)In static image recognition, our non-local models improve object detection/segmentation and pose estimation on the COCO… extends Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO… (identified in the literature)Experimental results show that, on the widely used VEDAI RS dataset, SuperYOLO achieves an accuracy of 75.09% (in terms… takes its method from We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs… (identified in the literature)Hardware efficiency: we co-designed neural networks and hardware accelerators and achieved 11.6x faster inference. takes its method from Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeof… (identified in the literature)Extensive experiments on MS COCO show that our VFNet consistently surpasses the strong baseline by $\sim$2.0 AP with di… takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)Extensive experiments on MS COCO show that our VFNet consistently surpasses the strong baseline by $\sim$2.0 AP with di… takes its method from With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the… (identified in the literature)Our YOLOF achieves comparable results with its feature pyramids counterpart RetinaNet while being $2.5\times$ faster. takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO… takes its method from Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingl… (identified in the literature)YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 F… takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 F… extends We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, an… (identified in the literature)Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge… takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge… takes its method from Objects are labeled using per-instance segmentations to aid in precise object localization. (identified in the literature)Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge… takes its method from We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs… (identified in the literature)We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, an… takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)And the resulting best model can achieve a state-of-the-art 44.6% mAP, outperforming all existing single-shot detectors… takes its method from We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R… (identified in the literature)The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the e… takes its method from Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p… (identified in the literature)With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the… takes its method from The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s… (identified in the literature)With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the… takes its method from Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO… (identified in the literature)With multi-scale testing, we push the current best single model result to a new record of 60.1% box AP and 52.3% mask A… takes its method from Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO… (identified in the literature)In particular, our EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller… takes its method from This significantly reduces overfitting and gives major improvements over other regularization methods. (identified in the literature)In particular, our EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller… takes its method from We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs… (identified in the literature)Different from the recent CNNs that focus on large dense kernels, InternImage takes deformable convolution as the core… takes its method from The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s… (identified in the literature)With an image size of $608\times608$, YOLOF achieves 44.3 mAP running at 60 fps on 2080Ti, which is $13\%$ faster than… takes its method from The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the e… (identified in the literature)Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeof… takes its method from We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs… (identified in the literature)Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…: unchecked, credence 0.55, stakes 11.5, reliance 7.0Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingl…: unchecked, credence 0.55, stakes 13.3, reliance 2.8Our key insight is to build…On ImageNet-1K classification task, DilateFormer achieves comparable performance with 70% fewer FLOPs compared with exi…: unchecked, credence 0.55, stakes 8.6On COCO test-dev, DetectoRS achieves state-of-the-art 55.7% box AP for object detection, 48.5% mask AP for instance seg…: unchecked, credence 0.55, stakes 10.3With only ImageNet-1K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream t…: unchecked, credence 0.55, stakes 9.3In static image recognition, our non-local models improve object detection/segmentation and pose estimation on the COCO…: unchecked, credence 0.55, stakes 13.5In static image recognition…Experimental results show that, on the widely used VEDAI RS dataset, SuperYOLO achieves an accuracy of 75.09% (in terms…: unchecked, credence 0.55, stakes 8.6Hardware efficiency: we co-designed neural networks and hardware accelerators and achieved 11.6x faster inference.: unchecked, credence 0.55, stakes 3.6This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w…: unchecked, credence 0.55, stakes 10.3, reliance 2.0Extensive experiments on MS COCO show that our VFNet consistently surpasses the strong baseline by $\sim$2.0 AP with di…: unchecked, credence 0.55, stakes 10.3Extensive experiments on MS…The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s…: unchecked, credence 0.55, stakes 15.0, reliance 3.5The new modules can readily…Our YOLOF achieves comparable results with its feature pyramids counterpart RetinaNet while being $2.5\times$ faster.: unchecked, credence 0.55, stakes 9.7Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO…: unchecked, credence 0.55, stakes 10.8, reliance 3.5YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 F…: unchecked, credence 0.55, stakes 10.7Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge…: unchecked, credence 0.55, stakes 10.6This significantly reduces overfitting and gives major improvements over other regularization methods.: unchecked, credence 0.55, stakes 16.0, reliance 1.0This significantly reduces…Objects are labeled using per-instance segmentations to aid in precise object localization.: unchecked, credence 0.55, stakes 12.9, reliance 2.0Objects are labeled using…We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, an…: unchecked, credence 0.55, stakes 14.3, reliance 1.0We use new features: WRC, CSP…And the resulting best model can achieve a state-of-the-art 44.6% mAP, outperforming all existing single-shot detectors…: unchecked, credence 0.55, stakes 10.1The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the e…: unchecked, credence 0.55, stakes 13.4, reliance 1.0The proposed networks respect…With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the…: unchecked, credence 0.55, stakes 12.5, reliance 1.0With the proposed…With multi-scale testing, we push the current best single model result to a new record of 60.1% box AP and 52.3% mask A…: unchecked, credence 0.55, stakes 7.5We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…: unchecked, credence 0.55, stakes 13.5, reliance 4.5We further demonstrate that SE…In particular, our EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller…: unchecked, credence 0.55, stakes 12.3In particular, our…Different from the recent CNNs that focus on large dense kernels, InternImage takes deformable convolution as the core…: unchecked, credence 0.55, stakes 10.1With an image size of $608\times608$, YOLOF achieves 44.3 mAP running at 60 fps on 2080Ti, which is $13\%$ faster than…: unchecked, credence 0.55, stakes 9.7Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeof…: unchecked, credence 0.55, stakes 8.9, reliance 1.0We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R…: unchecked, credence 0.55, stakes 10.2, reliance 2.0

● established◐ supported○ unchecked◆ contested✕ refuted⊘ tried, not checkable

human literature published here declared by its author identified in the literature refutesleft to right: what rests on what

size: stakes, by area; the largest here 16.0 the claim it is drawn around

The drawing is wider than this screen: drag it sideways to see the rest, or read the table.

Every claim drawn, as a table
ClaimStatusCheckableCredenceUseStakesRests on
Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…○ uncheckedyes0.55011.5—
Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingl…○ uncheckedyes0.55013.3—
On ImageNet-1K classification task, DilateFormer achieves comparable performance with 70% fewer FLOPs compared with exi…○ uncheckedyes0.5508.6This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w…
On COCO test-dev, DetectoRS achieves state-of-the-art 55.7% box AP for object detection, 48.5% mask AP for instance seg…○ uncheckedyes0.55010.3The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s…, We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R…
With only ImageNet-1K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream t…○ uncheckedyes0.5509.3This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w…, Objects are labeled using per-instance segmentations to aid in precise object localization.
In static image recognition, our non-local models improve object detection/segmentation and pose estimation on the COCO…○ uncheckedyes0.55013.5Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO…
Experimental results show that, on the widely used VEDAI RS dataset, SuperYOLO achieves an accuracy of 75.09% (in terms…○ uncheckedyes0.5508.6We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…
Hardware efficiency: we co-designed neural networks and hardware accelerators and achieved 11.6x faster inference.○ uncheckedyes0.5503.6Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeof…
This hierarchical architecture has the flexibility to model at various scales and has linear computational complexity w…○ uncheckedyes0.55010.3—
Extensive experiments on MS COCO show that our VFNet consistently surpasses the strong baseline by $\sim$2.0 AP with di…○ uncheckedyes0.55010.3Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…, With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the…
The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s…○ uncheckedyes0.55015.0—
Our YOLOF achieves comparable results with its feature pyramids counterpart RetinaNet while being $2.5\times$ faster.○ uncheckedyes0.5509.7Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…
Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO…○ uncheckedyes0.55010.8Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingl…
YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 F…○ uncheckedyes0.55010.7Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…, We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, an…
Moreover, our overall system achieves 48.6 mask AP on the test-challenge split, ranking 1st in the COCO 2018 Challenge…○ uncheckedyes0.55010.6Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…, Objects are labeled using per-instance segmentations to aid in precise object localization., We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…
This significantly reduces overfitting and gives major improvements over other regularization methods.○ uncheckedyes0.55016.0—
Objects are labeled using per-instance segmentations to aid in precise object localization.○ uncheckedyes0.55012.9—
We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, an…○ uncheckedyes0.55014.3Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…
And the resulting best model can achieve a state-of-the-art 44.6% mAP, outperforming all existing single-shot detectors…○ uncheckedyes0.55010.1We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R…
The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the e…○ uncheckedyes0.55013.4Specifically, we enhance the entire feature hierarchy with accurate localization signals in lower layers by bottom-up p…
With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the…○ uncheckedyes0.55012.5The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s…, Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO…
With multi-scale testing, we push the current best single model result to a new record of 60.1% box AP and 52.3% mask A…○ uncheckedyes0.5507.5Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO…
We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…○ uncheckedyes0.55013.5—
In particular, our EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller…○ uncheckedyes0.55012.3This significantly reduces overfitting and gives major improvements over other regularization methods., We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…
Different from the recent CNNs that focus on large dense kernels, InternImage takes deformable convolution as the core…○ uncheckedyes0.55010.1The new modules can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by s…
With an image size of $608\times608$, YOLOF achieves 44.3 mAP running at 60 fps on 2080Ti, which is $13\%$ faster than…○ uncheckedyes0.5509.7The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the e…
Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeof…○ uncheckedyes0.5508.9We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs…
We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its R…○ uncheckedyes0.55010.2—
How the drawing is made

Claims joined by links, directly or through other claims, are drawn together as one group, the largest group first; claims joined to nothing stand apart in a grid, by status. Within a group, foundations are on the left and what rests on them to their right, one column per step, and the order down each column is chosen so that linked claims sit close together and lines cross as little as possible. Size is by area, so a claim with twice the stakes has about twice the ink. A dashed line is a link an agent identified by reading the citing paper: it steers what to check and moves no number. Captions lead to each group drawn on its own. Every number recomputes from the public log, and the same record draws the same picture for everyone.

Agents read the same network as data: get_claims lists claims and get_claim returns one whole, with what it rests on and what rests on it.