doc_id stringclasses 80
values | sentence stringlengths 22 960 | ner listlengths 0 24 | rel listlengths 0 14 | raw_prediction stringlengths 22 878 |
|---|---|---|---|---|
51923817 | We propose CornerNet , a new approach to object detection where we detect an object bounding box as a pair of keypoints , the top - left corner and the bottom - right corner , using a single convolution neural network . | [
[
"CornerNet",
"Method"
],
[
"object detection",
"Task"
],
[
"convolution neural network",
"Method"
]
] | [
[
"CornerNet",
"Used-For",
"object detection"
]
] | ner: CornerNet|Method ; object detection|Task ; convolution neural network|Method || rel: CornerNet|Used-For|object detection |
51923817 | Experiments show that CornerNet achieves a 4 2 . 2 % AP on MS COCO , outperforming all existing one - stage detectors . | [
[
"CornerNet",
"Method"
],
[
"MS COCO",
"Dataset"
]
] | [
[
"CornerNet",
"Evaluated-With",
"MS COCO"
]
] | ner: CornerNet|Method ; MS COCO|Dataset || rel: CornerNet|Evaluated-With|MS COCO |
51923817 | Object detectors based on convolutional neural networks ( ConvNets ) ( Krizhevsky et al. , 2 0 1 2 ; Simonyan and Zisserman , 2 0 1 4 ; He et al. , 2 0 1 6 ) have achieved state - of - the - art results on various challenging benchmarks ( Lin et al. , 2 0 1 4 ; Deng et al. , 2 0 0 9 ; Everingham et al. , 2 0 1 5 ) . | [
[
"convolutional neural networks",
"Method"
],
[
"ConvNets",
"Method"
]
] | [
[
"ConvNets",
"Synonym-Of",
"convolutional neural networks"
]
] | ner: convolutional neural networks|Method ; ConvNets|Method || rel: ConvNets|Synonym-Of|convolutional neural networks |
51923817 | In this paper we introduce CornerNet , a new onestage approach to object detection that does away with anchor boxes . | [
[
"CornerNet",
"Method"
],
[
"object detection",
"Task"
]
] | [
[
"CornerNet",
"Used-For",
"object detection"
]
] | ner: CornerNet|Method ; object detection|Task || rel: CornerNet|Used-For|object detection |
51923817 | We use a single convolutional network to predict a heatmap for the top - left corners of all instances of the same object category , a heatmap for all bottomright corners , and an embedding vector for each detected corner . | [
[
"convolutional network",
"Method"
],
[
"predict a heatmap",
"Task"
]
] | [
[
"convolutional network",
"Used-For",
"predict a heatmap"
]
] | ner: convolutional network|Method ; predict a heatmap|Task || rel: convolutional network|Used-For|predict a heatmap |
51923817 | A convolutional network outputs a heatmap for all top - left corners , a heatmap for all bottom - right corners , and an embedding vector for each detected corner . | [
[
"convolutional network",
"Method"
]
] | [] | ner: convolutional network|Method || rel: none |
51923817 | Another novel component of CornerNet is corner pooling , a new type of pooling layer that helps a convolutional network better localize corners of bounding boxes . | [
[
"CornerNet",
"Method"
],
[
"corner pooling",
"Method"
],
[
"pooling layer",
"Method"
],
[
"convolutional network",
"Method"
]
] | [
[
"corner pooling",
"Part-Of",
"CornerNet"
],
[
"convolutional network",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; corner pooling|Method ; pooling layer|Method ; convolutional network|Method || rel: corner pooling|Part-Of|CornerNet ; convolutional network|Part-Of|CornerNet |
51923817 | We demonstrate the effectiveness of CornerNet on MS COCO ( Lin et al. , 2 0 1 4 ) . | [
[
"CornerNet",
"Method"
],
[
"MS COCO",
"Dataset"
]
] | [
[
"CornerNet",
"Evaluated-With",
"MS COCO"
]
] | ner: CornerNet|Method ; MS COCO|Dataset || rel: CornerNet|Evaluated-With|MS COCO |
51923817 | In addition , through ablation studies we show that corner pooling is critical to the superior performance of CornerNet . | [
[
"corner pooling",
"Method"
],
[
"CornerNet",
"Method"
]
] | [
[
"corner pooling",
"Part-Of",
"CornerNet"
]
] | ner: corner pooling|Method ; CornerNet|Method || rel: corner pooling|Part-Of|CornerNet |
51923817 | Code is available at https://github.com/ princeton - vl/CornerNet . Two - stage approach was first introduced and popularized by R - CNN ( Girshick et al. , 2 0 1 4 ) . | [
[
"R - CNN",
"Method"
]
] | [] | ner: R - CNN|Method || rel: none |
51923817 | Later , SPP ( He et al. , 2 0 1 4 ) and Fast - RCNN ( Girshick , 2 0 1 5 ) improve R - CNN by designing a special pooling layer that pools each region from feature maps instead . | [
[
"SPP",
"Method"
],
[
"Fast - RCNN",
"Method"
],
[
"R - CNN",
"Method"
]
] | [
[
"SPP",
"Compare-With",
"Fast - RCNN"
],
[
"Fast - RCNN",
"Compare-With",
"R - CNN"
]
] | ner: SPP|Method ; Fast - RCNN|Method ; R - CNN|Method || rel: SPP|Compare-With|Fast - RCNN ; Fast - RCNN|Compare-With|R - CNN |
51923817 | Faster - RCNN ( Ren et al. , 2 0 1 5 ) does away low level proposal algorithms by introducing a region proposal network ( RPN ) , which generates proposals from a set of pre - determined candidate boxes , usually known as anchor boxes . | [
[
"Faster - RCNN",
"Method"
],
[
"region proposal network",
"Method"
],
[
"RPN",
"Method"
]
] | [
[
"region proposal network",
"Part-Of",
"Faster - RCNN"
],
[
"RPN",
"Synonym-Of",
"region proposal network"
]
] | ner: Faster - RCNN|Method ; region proposal network|Method ; RPN|Method || rel: region proposal network|Part-Of|Faster - RCNN ; RPN|Synonym-Of|region proposal network |
51923817 | R - FCN ( Dai et al. , 2 0 1 6 ) further improves the efficiency of Faster - RCNN by replacing the fully connected sub - detection network with a fully convolutional subdetection network . | [
[
"R - FCN",
"Method"
],
[
"Faster - RCNN",
"Method"
],
[
"fully connected sub - detection network",
"Method"
],
[
"fully convolutional subdetection network",
"Method"
]
] | [
[
"R - FCN",
"Compare-With",
"Faster - RCNN"
],
[
"fully connected sub - detection network",
"Part-Of",
"Faster - RCNN"
],
[
"fully convolutional subdetection network",
"Part-Of",
"Faster - RCNN"
]
] | ner: R - FCN|Method ; Faster - RCNN|Method ; fully connected sub - detection network|Method ; fully convolutional subdetection network|Method || rel: R - FCN|Compare-With|Faster - RCNN ; fully connected sub - detection network|Part-Of|Faster - RCNN ; fully convolutional subdetection network|Part-Of|Faster - RCNN |
51923817 | On the other hand , YOLO and SSD ( Liu et al. , 2 0 1 6 ) have popularized the one - stage approach , which removes the RoI pooling step and detects objects in a single network . | [
[
"YOLO",
"Method"
],
[
"SSD",
"Method"
],
[
"RoI pooling",
"Method"
]
] | [] | ner: YOLO|Method ; SSD|Method ; RoI pooling|Method || rel: none |
51923817 | YOLO predicts bounding box coordinates directly from an image , and is later improved in YOLO 9 0 0 0 by switching to anchor boxes . | [
[
"YOLO",
"Method"
],
[
"YOLO 9 0 0 0",
"Dataset"
]
] | [] | ner: YOLO|Method ; YOLO 9 0 0 0|Dataset || rel: none |
51923817 | DSSD and RON ( Kong et al. , 2 0 1 7 ) adopt networks similar to the hourglass network ( Newell et al. , 2 0 1 6 ) , enabling them to combine low - level and high - level features via skip connections to predict bounding boxes more accurately . | [
[
"DSSD",
"Method"
],
[
"RON",
"Method"
],
[
"hourglass network",
"Method"
],
[
"skip connections",
"Method"
]
] | [
[
"skip connections",
"Part-Of",
"DSSD"
],
[
"skip connections",
"Part-Of",
"RON"
]
] | ner: DSSD|Method ; RON|Method ; hourglass network|Method ; skip connections|Method || rel: skip connections|Part-Of|DSSD ; skip connections|Part-Of|RON |
51923817 | Second , DeNet selects features at manually determined locations relative to a region for classification , while our approach does not require any feature selection step . | [
[
"DeNet",
"Method"
],
[
"classification",
"Task"
]
] | [
[
"DeNet",
"Used-For",
"classification"
]
] | ner: DeNet|Method ; classification|Task || rel: DeNet|Used-For|classification |
51923817 | Point Linking Network ( PLN ) is an one - stage detector without anchor boxes . | [
[
"Point Linking Network",
"Method"
],
[
"PLN",
"Method"
]
] | [
[
"PLN",
"Synonym-Of",
"Point Linking Network"
]
] | ner: Point Linking Network|Method ; PLN|Method || rel: PLN|Synonym-Of|Point Linking Network |
51923817 | CornerNet is very different from PLN . | [
[
"CornerNet",
"Method"
],
[
"PLN",
"Method"
]
] | [
[
"CornerNet",
"Compare-With",
"PLN"
]
] | ner: CornerNet|Method ; PLN|Method || rel: CornerNet|Compare-With|PLN |
51923817 | Second , CornerNet uses corner pooling to better localize the corners . | [
[
"CornerNet",
"Method"
],
[
"corner pooling",
"Method"
]
] | [
[
"corner pooling",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; corner pooling|Method || rel: corner pooling|Part-Of|CornerNet |
51923817 | We also significantly modify the hourglass architecture and add our novel variant of focal loss to help better train the network . In CornerNet , we detect an object as a pair of keypointsthe top - left corner and bottom - right corner of the bounding box . | [
[
"focal loss",
"Method"
],
[
"CornerNet",
"Method"
]
] | [] | ner: focal loss|Method ; CornerNet|Method || rel: none |
51923817 | We use the hourglass network ( Newell et al. , 2 0 1 6 ) as the backbone network of CornerNet . | [
[
"hourglass network",
"Method"
],
[
"CornerNet",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
]
] | ner: hourglass network|Method ; CornerNet|Method || rel: hourglass network|Part-Of|CornerNet |
51923817 | Each module has its own corner pooling module to pool features from the hourglass network before predicting the heatmaps , embeddings and offsets . | [
[
"corner pooling",
"Method"
],
[
"hourglass network",
"Method"
]
] | [] | ner: corner pooling|Method ; hourglass network|Method || rel: none |
51923817 | In this modified residual block , we replace the first 3 × 3 convolution module with a corner pooling module , which first processes the features from the backbone network by two 3 × 3 convolution modules 1 with 1 2 8 channels and then applies a corner pooling layer . | [
[
"residual block",
"Method"
],
[
"3 3 convolution",
"Method"
],
[
"corner pooling",
"Method"
],
[
"3 3 convolution",
"Method"
],
[
"corner pooling",
"Method"
]
] | [] | ner: residual block|Method ; 3 3 convolution|Method ; corner pooling|Method ; 3 3 convolution|Method ; corner pooling|Method || rel: none |
51923817 | Following the design of a residual block , we then feed the pooled features into a 3 × 3 Conv - BN layer with 2 5 6 channels and add back the projection shortcut . | [
[
"residual block",
"Method"
],
[
"3 3 Conv - BN",
"Method"
]
] | [] | ner: residual block|Method ; 3 3 Conv - BN|Method || rel: none |
51923817 | The modified residual block is followed by a 3 × 3 convolution module with 2 5 6 channels , and 3 Conv - ReLU - Conv layers to produce the heatmaps , embeddings and offsets . | [
[
"residual block",
"Method"
],
[
"3 3 convolution module",
"Method"
],
[
"Conv - ReLU - Conv",
"Method"
]
] | [] | ner: residual block|Method ; 3 3 convolution module|Method ; Conv - ReLU - Conv|Method || rel: none |
51923817 | CornerNet uses the hourglass network ( Newell et al. , 2 0 1 6 ) as its backbone network . | [
[
"CornerNet",
"Method"
],
[
"hourglass network",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; hourglass network|Method || rel: hourglass network|Part-Of|CornerNet |
51923817 | The hourglass network was first introduced for the human pose estimation task . | [
[
"hourglass network",
"Method"
],
[
"human pose estimation",
"Task"
]
] | [
[
"hourglass network",
"Used-For",
"human pose estimation"
]
] | ner: hourglass network|Method ; human pose estimation|Task || rel: hourglass network|Used-For|human pose estimation |
51923817 | An hourglass module first downsamples the input features by a series of convolution and max pooling layers . | [
[
"hourglass module",
"Method"
],
[
"convolution",
"Method"
],
[
"max pooling",
"Method"
]
] | [] | ner: hourglass module|Method ; convolution|Method ; max pooling|Method || rel: none |
51923817 | These properties make the hourglass network an ideal choice for object detection as well . | [
[
"hourglass network",
"Method"
],
[
"object detection",
"Task"
]
] | [
[
"hourglass network",
"Used-For",
"object detection"
]
] | ner: hourglass network|Method ; object detection|Task || rel: hourglass network|Used-For|object detection |
51923817 | Our hourglass network consists of two hourglasses , and we make some modifications to the architecture of the hourglass module . | [
[
"hourglass network",
"Method"
],
[
"hourglasses",
"Method"
]
] | [] | ner: hourglass network|Method ; hourglasses|Method || rel: none |
51923817 | We scan from right to left for the horizontal max - pooling and from bottom to top for the vertical max - pooling . | [
[
"horizontal max - pooling",
"Method"
],
[
"vertical max - pooling",
"Method"
]
] | [] | ner: horizontal max - pooling|Method ; vertical max - pooling|Method || rel: none |
51923817 | Fig. 7 The prediction module starts with a modified residual block , in which we replace the first convolution module with our corner pooling module . | [
[
"residual block",
"Method"
],
[
"convolution",
"Method"
],
[
"corner pooling",
"Method"
]
] | [] | ner: residual block|Method ; convolution|Method ; corner pooling|Method || rel: none |
51923817 | We apply a 1 × 1 Conv - BN module to both the input and output of the first hourglass module . | [
[
"1 1 Conv - BN",
"Method"
],
[
"hourglass",
"Method"
]
] | [
[
"1 1 Conv - BN",
"Part-Of",
"hourglass"
]
] | ner: 1 1 Conv - BN|Method ; hourglass|Method || rel: 1 1 Conv - BN|Part-Of|hourglass |
51923817 | We then merge them by element - wise addition followed by a ReLU and a residual block with 2 5 6 channels , which is then used as the input to the second hourglass module . | [
[
"ReLU",
"Method"
],
[
"residual block",
"Method"
],
[
"hourglass module",
"Method"
]
] | [] | ner: ReLU|Method ; residual block|Method ; hourglass module|Method || rel: none |
51923817 | To reduce overfitting , we adopt standard data augmentation techniques including random horizontal flipping , random scaling , random cropping and random color jittering , which includes adjusting the brightness , saturation and contrast of an image . | [
[
"data augmentation",
"Method"
],
[
"random horizontal flipping",
"Method"
],
[
"random scaling",
"Method"
],
[
"random cropping",
"Method"
],
[
"random color jittering",
"Method"
]
] | [
[
"random horizontal flipping",
"SubClass-Of",
"data augmentation"
],
[
"random scaling",
"SubClass-Of",
"data augmentation"
],
[
"random cropping",
"SubClass-Of",
"data augmentation"
]
] | ner: data augmentation|Method ; random horizontal flipping|Method ; random scaling|Method ; random cropping|Method ; random color jittering|Method || rel: random horizontal flipping|SubClass-Of|data augmentation ; random scaling|SubClass-Of|data augmentation ; random cropping|SubClass-Of|data augmentation |
51923817 | We first apply non - maximal suppression ( NMS ) by using a 3 × 3 max pooling layer on the corner heatmaps . | [
[
"non - maximal suppression",
"Method"
],
[
"NMS",
"Method"
],
[
"3 3 max pooling",
"Method"
]
] | [
[
"NMS",
"Synonym-Of",
"non - maximal suppression"
],
[
"3 3 max pooling",
"Part-Of",
"non - maximal suppression"
]
] | ner: non - maximal suppression|Method ; NMS|Method ; 3 3 max pooling|Method || rel: NMS|Synonym-Of|non - maximal suppression ; 3 3 max pooling|Part-Of|non - maximal suppression |
51923817 | We evaluate CornerNet on the very challenging MS COCO dataset ( Lin et al. , 2 0 1 4 ) . | [
[
"CornerNet",
"Method"
],
[
"MS COCO",
"Dataset"
]
] | [
[
"CornerNet",
"Evaluated-With",
"MS COCO"
]
] | ner: CornerNet|Method ; MS COCO|Dataset || rel: CornerNet|Evaluated-With|MS COCO |
51923817 | MS COCO uses average precisions ( APs ) at different IoUs and APs for different object sizes as the main evaluation metrics . Corner pooling is a key component of CornerNet . | [
[
"MS COCO",
"Dataset"
],
[
"Corner pooling",
"Method"
],
[
"CornerNet",
"Method"
]
] | [
[
"Corner pooling",
"Part-Of",
"MS COCO"
]
] | ner: MS COCO|Dataset ; Corner pooling|Method ; CornerNet|Method || rel: Corner pooling|Part-Of|MS COCO |
51923817 | Corner pooling gives similar improvement to corners at different quadrants , show that corner pooling is effective and stable over both small and large areas . | [
[
"Corner pooling",
"Method"
],
[
"corner pooling",
"Method"
]
] | [] | ner: Corner pooling|Method ; corner pooling|Method || rel: none |
51923817 | CornerNet uses the hourglass network ( Newell et al. , 2 0 1 6 ) as its backbone network . | [
[
"CornerNet",
"Method"
],
[
"hourglass network",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; hourglass network|Method || rel: hourglass network|Part-Of|CornerNet |
51923817 | Since the hourglass network is not commonly used in other state - of - the - art detectors , we perform an experiment to study the contribution of the hourglass network in CornerNet . | [
[
"hourglass network",
"Method"
],
[
"hourglass network",
"Method"
],
[
"CornerNet",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
]
] | ner: hourglass network|Method ; hourglass network|Method ; CornerNet|Method || rel: hourglass network|Part-Of|CornerNet |
51923817 | We train a CornerNet in which we replace the hourglass network with FPN ( w/ ResNet - 1 0 1 ) , which is more commonly used in state - of - the - art object detectors . | [
[
"CornerNet",
"Method"
],
[
"hourglass network",
"Method"
],
[
"FPN",
"Method"
],
[
"ResNet - 1 0 1",
"Method"
],
[
"object detectors",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
],
[
"FPN",
"Part-Of",
"CornerNet"
],
[
"ResNet - 1 0 1",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; hourglass network|Method ; FPN|Method ; ResNet - 1 0 1|Method ; object detectors|Method || rel: hourglass network|Part-Of|CornerNet ; FPN|Part-Of|CornerNet ; ResNet - 1 0 1|Part-Of|CornerNet |
51923817 | Tab . 4 shows that CornerNet with hourglass network outperforms CornerNet with FPN by 8. 2 % AP , and the anchor box based detector with hourglass network by 5. 5 % AP . | [
[
"CornerNet",
"Method"
],
[
"hourglass network",
"Method"
],
[
"CornerNet",
"Method"
],
[
"FPN",
"Method"
],
[
"anchor box based detector",
"Method"
],
[
"hourglass network",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
],
[
"hourglass network",
"Part-Of",
"CornerNet"
],
[
"anchor box based detector",
"Part-Of",
"CornerNet"
]
] | ner: CornerNet|Method ; hourglass network|Method ; CornerNet|Method ; FPN|Method ; anchor box based detector|Method ; hourglass network|Method || rel: hourglass network|Part-Of|CornerNet ; hourglass network|Part-Of|CornerNet ; anchor box based detector|Part-Of|CornerNet |
51923817 | The results suggest that the choice of the backbone network is important and the hourglass network is crucial to the performance of CornerNet . A good detector should predict high quality bounding boxes that cover objects tightly . | [
[
"hourglass network",
"Method"
],
[
"CornerNet",
"Method"
]
] | [
[
"hourglass network",
"Part-Of",
"CornerNet"
]
] | ner: hourglass network|Method ; CornerNet|Method || rel: hourglass network|Part-Of|CornerNet |
51923817 | To understand the quality of the bounding boxes predicted by CornerNet , we evaluate the performance of CornerNet at multiple IoU thresholds , and compare the results with other state - of - the - art detectors , including RetinaNet , Cascade R - CNN ( Cai and Vasconcelos , 2 0 1 7 ) and IoU - Net ( Jiang et al. , 2 0 ... | [
[
"CornerNet",
"Method"
],
[
"CornerNet",
"Method"
],
[
"RetinaNet",
"Method"
],
[
"Cascade R - CNN",
"Method"
],
[
"IoU - Net",
"Method"
]
] | [
[
"CornerNet",
"Compare-With",
"RetinaNet"
],
[
"CornerNet",
"Compare-With",
"Cascade R - CNN"
],
[
"CornerNet",
"Compare-With",
"IoU - Net"
]
] | ner: CornerNet|Method ; CornerNet|Method ; RetinaNet|Method ; Cascade R - CNN|Method ; IoU - Net|Method || rel: CornerNet|Compare-With|RetinaNet ; CornerNet|Compare-With|Cascade R - CNN ; CornerNet|Compare-With|IoU - Net |
51923817 | Tab . 5 shows that CornerNet achieves a much higher AP at 0. 9 IoU than other detectors , outperforming Cascade R - CNN + IoU - Net by 3. 9 % , Cascade R - CNN by 7. 6 % and RetinaNet 2 by 7. 3 % . | [
[
"CornerNet",
"Method"
],
[
"Cascade R - CNN + IoU - Net",
"Method"
],
[
"Cascade R - CNN",
"Method"
],
[
"RetinaNet",
"Method"
]
] | [
[
"CornerNet",
"Compare-With",
"Cascade R - CNN + IoU - Net"
],
[
"CornerNet",
"Compare-With",
"Cascade R - CNN"
],
[
"CornerNet",
"Compare-With",
"RetinaNet"
]
] | ner: CornerNet|Method ; Cascade R - CNN + IoU - Net|Method ; Cascade R - CNN|Method ; RetinaNet|Method || rel: CornerNet|Compare-With|Cascade R - CNN + IoU - Net ; CornerNet|Compare-With|Cascade R - CNN ; CornerNet|Compare-With|RetinaNet |
51923817 | CornerNet simultaneously outputs heatmaps , offsets , and embeddings , all of which affect detection performance . | [
[
"CornerNet",
"Dataset"
],
[
"detection",
"Task"
]
] | [
[
"CornerNet",
"Benchmark-For",
"detection"
]
] | ner: CornerNet|Dataset ; detection|Task || rel: CornerNet|Benchmark-For|detection |
51923817 | We compare CornerNet with other state - of - the - art detectors on MS COCO test - dev ( Tab . 7 ) . | [
[
"CornerNet",
"Method"
],
[
"MS COCO test - dev",
"Dataset"
]
] | [
[
"CornerNet",
"Evaluated-With",
"MS COCO test - dev"
]
] | ner: CornerNet|Method ; MS COCO test - dev|Dataset || rel: CornerNet|Evaluated-With|MS COCO test - dev |
51923817 | We have presented CornerNet , a new approach to object detection that detects bounding boxes as pairs of corners . | [
[
"CornerNet",
"Method"
],
[
"object detection",
"Task"
]
] | [
[
"CornerNet",
"Used-For",
"object detection"
]
] | ner: CornerNet|Method ; object detection|Task || rel: CornerNet|Used-For|object detection |
51923817 | We evaluate CornerNet on MS COCO and demonstrate competitive results . | [
[
"CornerNet",
"Method"
],
[
"MS COCO",
"Dataset"
]
] | [
[
"CornerNet",
"Evaluated-With",
"MS COCO"
]
] | ner: CornerNet|Method ; MS COCO|Dataset || rel: CornerNet|Evaluated-With|MS COCO |
104291983 | High - resolution representation learning plays an essential role in many vision problems , e.g. , pose estimation and semantic segmentation . | [
[
"High - resolution representation learning",
"Method"
],
[
"pose estimation",
"Task"
],
[
"semantic segmentation",
"Task"
]
] | [
[
"High - resolution representation learning",
"Used-For",
"pose estimation"
],
[
"High - resolution representation learning",
"Used-For",
"semantic segmentation"
]
] | ner: High - resolution representation learning|Method ; pose estimation|Task ; semantic segmentation|Task || rel: High - resolution representation learning|Used-For|pose estimation ; High - resolution representation learning|Used-For|semantic segmentation |
104291983 | The high - resolution network (HRNet)~\cite{SunXLW 1 9 } , recently developed for human pose estimation , maintains high - resolution representations through the whole process by connecting high - to - low resolution convolutions in \emph{parallel } and produces strong high - resolution representations by repeatedly co... | [
[
"high - resolution network",
"Method"
],
[
"HRNet",
"Method"
],
[
"SunXLW 1 9",
"Method"
],
[
"human pose estimation",
"Task"
],
[
"high - to - low resolution convolutions",
"Method"
],
[
"parallel convolutions",
"Method"
]
] | [
[
"HRNet",
"Synonym-Of",
"high - resolution network"
],
[
"high - resolution network",
"Used-For",
"human pose estimation"
],
[
"high - to - low resolution convolutions",
"Part-Of",
"high - resolution network"
],
[
"high - to - low resolution convolutions",
"P... | ner: high - resolution network|Method ; HRNet|Method ; SunXLW 1 9 |Method ; human pose estimation|Task ; high - to - low resolution convolutions|Method ; parallel convolutions|Method || rel: HRNet|Synonym-Of|high - resolution network ; high - resolution network|Used-For|human pose estimation ; high - to - low resolutio... |
104291983 | We show top results in semantic segmentation on Cityscapes , LIP , and PASCAL Context , and facial landmark detection on AFLW , COFW , $ 3 0 0 $W , and WFLW . | [
[
"semantic segmentation",
"Task"
],
[
"Cityscapes",
"Dataset"
],
[
"LIP",
"Dataset"
],
[
"PASCAL Context",
"Dataset"
],
[
"facial landmark detection",
"Task"
],
[
"AFLW",
"Dataset"
],
[
"COFW",
"Dataset"
],
[
"$ 3 0 0 $W",
... | [
[
"Cityscapes",
"Benchmark-For",
"semantic segmentation"
],
[
"LIP",
"Benchmark-For",
"semantic segmentation"
],
[
"PASCAL Context",
"Benchmark-For",
"semantic segmentation"
],
[
"AFLW",
"Benchmark-For",
"semantic segmentation"
],
[
"COFW",
"Be... | ner: semantic segmentation|Task ; Cityscapes|Dataset ; LIP|Dataset ; PASCAL Context|Dataset ; facial landmark detection|Task ; AFLW|Dataset ; COFW|Dataset ; $ 3 0 0 $W|Dataset ; WFLW|Dataset || rel: Cityscapes|Benchmark-For|semantic segmentation ; LIP|Benchmark-For|semantic segmentation ; PASCAL Context|Benchmark-For|s... |
104291983 | In addition , we build a multi - level representation from the high - resolution representation and apply it to the Faster R - CNN object detection framework and the extended frameworks . | [
[
"Faster R - CNN",
"Method"
],
[
"object detection",
"Task"
]
] | [
[
"Faster R - CNN",
"Used-For",
"object detection"
]
] | ner: Faster R - CNN|Method ; object detection|Task || rel: Faster R - CNN|Used-For|object detection |
104291983 | The proposed approach achieves superior results to existing single - model networks on COCO object detection . | [
[
"COCO object detection",
"Dataset"
]
] | [] | ner: COCO object detection|Dataset || rel: none |
104291983 | There are two main kinds of representations : low - resolution representations that are mainly for image classification , and high - resolution representations that are essential for many other vision problems , e.g. , semantic segmentation , object detection , human pose estimation , etc . | [
[
"image classification",
"Task"
],
[
"semantic segmentation",
"Task"
],
[
"object detection",
"Task"
],
[
"human pose estimation",
"Task"
]
] | [] | ner: image classification|Task ; semantic segmentation|Task ; object detection|Task ; human pose estimation|Task || rel: none |
104291983 | One is to recover high - resolution representations from low - resolution representations outputted by a network ( e.g. , ResNet ) and optionally intermediate medium - resolution representations , e.g. , Hourglass [ 7 2 ] , SegNet [ 2 ] , DeconvNet [ 7 4 ] , U - Net [ 8 3 ] , and encoderdecoder [ 7 7 ] . | [
[
"ResNet",
"Method"
],
[
"Hourglass",
"Method"
],
[
"SegNet",
"Method"
],
[
"DeconvNet",
"Method"
],
[
"U - Net",
"Method"
],
[
"encoderdecoder",
"Method"
]
] | [] | ner: ResNet|Method ; Hourglass|Method ; SegNet|Method ; DeconvNet|Method ; U - Net|Method ; encoderdecoder|Method || rel: none |
104291983 | In addition , dilated convolutions are used to replace some strided convolutions and associated regular convolutions in classification networks to compute medium - resolution representations [ 1 3 , 1 2 6 ] . | [
[
"dilated convolutions",
"Method"
],
[
"strided convolutions",
"Method"
],
[
"convolutions",
"Method"
],
[
"classification networks",
"Method"
]
] | [
[
"dilated convolutions",
"Part-Of",
"classification networks"
],
[
"strided convolutions",
"Part-Of",
"classification networks"
],
[
"convolutions",
"Part-Of",
"classification networks"
]
] | ner: dilated convolutions|Method ; strided convolutions|Method ; convolutions|Method ; classification networks|Method || rel: dilated convolutions|Part-Of|classification networks ; strided convolutions|Part-Of|classification networks ; convolutions|Part-Of|classification networks |
104291983 | We go along the research line of maintaining highresolution representations and further study the highresolution network ( HRNet ) , which is initially developed for human pose estimation [ 9 1 ] , for a broad range of vision tasks . | [
[
"highresolution network",
"Method"
],
[
"HRNet",
"Method"
],
[
"human pose estimation",
"Task"
]
] | [
[
"HRNet",
"Synonym-Of",
"highresolution network"
],
[
"highresolution network",
"Used-For",
"human pose estimation"
]
] | ner: highresolution network|Method ; HRNet|Method ; human pose estimation|Task || rel: HRNet|Synonym-Of|highresolution network ; highresolution network|Used-For|human pose estimation |
104291983 | An HRNet maintains high - resolution representations by connecting high - to - low resolution convolutions in parallel and repeatedly conducting multi - scale fusions across parallel convolutions . | [
[
"HRNet",
"Method"
],
[
"high - to - low resolution convolutions",
"Method"
],
[
"parallel convolutions",
"Method"
]
] | [
[
"high - to - low resolution convolutions",
"Part-Of",
"HRNet"
]
] | ner: HRNet|Method ; high - to - low resolution convolutions|Method ; parallel convolutions|Method || rel: high - to - low resolution convolutions|Part-Of|HRNet |
104291983 | We apply our proposed network to semantic segmentation/facial landmark detection through estimating segmentation maps/facial landmark heatmaps from the output high - The 2nd ( 3rd , 4th ) stage repeats two - resolution ( three - resolution , four - resolution ) blocks . | [
[
"semantic segmentation/facial landmark detection",
"Task"
],
[
"segmentation maps/facial landmark heatmaps",
"Method"
]
] | [
[
"segmentation maps/facial landmark heatmaps",
"Used-For",
"semantic segmentation/facial landmark detection"
]
] | ner: semantic segmentation/facial landmark detection|Task ; segmentation maps/facial landmark heatmaps|Method || rel: segmentation maps/facial landmark heatmaps|Used-For|semantic segmentation/facial landmark detection |
104291983 | In semantic segmentation , the proposed approach achieves state - of - the - art results on PAS - CAL Context , Cityscapes , and LIP with similar model sizes and lower computation complexity . | [
[
"semantic segmentation",
"Task"
],
[
"PAS - CAL Context",
"Dataset"
],
[
"Cityscapes",
"Dataset"
],
[
"LIP",
"Dataset"
]
] | [
[
"PAS - CAL Context",
"Benchmark-For",
"semantic segmentation"
]
] | ner: semantic segmentation|Task ; PAS - CAL Context|Dataset ; Cityscapes|Dataset ; LIP|Dataset || rel: PAS - CAL Context|Benchmark-For|semantic segmentation |
104291983 | In facial landmark detection , our approach achieves overall best results on four standard datasets : AFLW , COFW , 3 0 0 W , and WFLW . | [
[
"facial landmark detection",
"Task"
],
[
"AFLW",
"Dataset"
],
[
"COFW",
"Dataset"
],
[
"3 0 0 W",
"Dataset"
],
[
"WFLW",
"Dataset"
]
] | [
[
"AFLW",
"Benchmark-For",
"facial landmark detection"
],
[
"COFW",
"Benchmark-For",
"facial landmark detection"
],
[
"3 0 0 W",
"Benchmark-For",
"facial landmark detection"
]
] | ner: facial landmark detection|Task ; AFLW|Dataset ; COFW|Dataset ; 3 0 0 W|Dataset ; WFLW|Dataset || rel: AFLW|Benchmark-For|facial landmark detection ; COFW|Benchmark-For|facial landmark detection ; 3 0 0 W|Benchmark-For|facial landmark detection |
104291983 | In addition , we construct a multi - level representation from the high - resolution representation , and apply it to the Faster R - CNN object detection framework and its extended frameworks , Mask R - CNN [ 3 8 ] and Cascade R - CNN [ 9 ] . | [
[
"Faster R - CNN",
"Method"
],
[
"object detection",
"Task"
],
[
"Mask R - CNN",
"Method"
],
[
"Cascade R - CNN",
"Method"
]
] | [
[
"Mask R - CNN",
"Used-For",
"object detection"
],
[
"Cascade R - CNN",
"Used-For",
"object detection"
]
] | ner: Faster R - CNN|Method ; object detection|Task ; Mask R - CNN|Method ; Cascade R - CNN|Method || rel: Mask R - CNN|Used-For|object detection ; Cascade R - CNN|Used-For|object detection |
104291983 | With single - scale training and testing , the proposed approach achieves better COCO object detection results than existing single - model methods . | [
[
"COCO object detection",
"Task"
]
] | [
[
"COCO object detection",
"Used-For",
"COCO object detection"
]
] | ner: COCO object detection|Task || rel: COCO object detection|Used-For|COCO object detection |
104291983 | Strong high - resolution representations play an essential role in pixel and region labeling problems , e.g. , semantic segmentation , human pose estimation , facial landmark detection , and object detection . | [
[
"semantic segmentation",
"Task"
],
[
"human pose estimation",
"Task"
],
[
"facial landmark detection",
"Task"
],
[
"object detection",
"Task"
]
] | [] | ner: semantic segmentation|Task ; human pose estimation|Task ; facial landmark detection|Task ; object detection|Task || rel: none |
104291983 | We review representation learning techniques developed mainly in the semantic segmentation , facial landmark detection [ 9 2 , 5 0 , 6 9 , 1 0 4 , 1 2 3 , 9 4 , 1 1 9 ] and object detection areas 1 , from low - resolution representation learning , high - resolution representation recovering , to high - resolution repre... | [
[
"representation learning",
"Method"
],
[
"semantic segmentation",
"Task"
],
[
"facial landmark detection",
"Task"
],
[
"object detection",
"Task"
],
[
"low - resolution representation learning",
"Method"
],
[
"high - resolution representation recover... | [
[
"representation learning",
"Used-For",
"semantic segmentation"
],
[
"representation learning",
"Used-For",
"facial landmark detection"
],
[
"representation learning",
"Used-For",
"object detection"
],
[
"low - resolution representation learning",
"Used-For",... | ner: representation learning|Method ; semantic segmentation|Task ; facial landmark detection|Task ; object detection|Task ; low - resolution representation learning|Method ; high - resolution representation recovering|Method ; high - resolution representation maintaining|Method || rel: representation learning|Used-For|... |
104291983 | The fullyconvolutional network ( FCN ) approaches [ 6 7 , 8 7 ] compute low - resolution representations by removing the fullyconnected layers in a classification network , and estimate from their coarse segmentation confidence maps . | [
[
"fullyconvolutional network",
"Method"
],
[
"FCN",
"Method"
],
[
"fullyconnected layers",
"Method"
],
[
"classification network",
"Method"
]
] | [
[
"FCN",
"Synonym-Of",
"fullyconvolutional network"
],
[
"fullyconnected layers",
"Part-Of",
"classification network"
]
] | ner: fullyconvolutional network|Method ; FCN|Method ; fullyconnected layers|Method ; classification network|Method || rel: FCN|Synonym-Of|fullyconvolutional network ; fullyconnected layers|Part-Of|classification network |
104291983 | The estimated segmentation maps are improved by combining the fine segmentation score maps estimated from intermediate low - level medium - resolution representations [ 6 7 ] , or iterating the processes [ 5 0 ] . | [
[
"average segmentation",
"Task"
],
[
"fine segmentation",
"Task"
]
] | [] | ner: average segmentation|Task ; fine segmentation|Task || rel: none |
104291983 | Similar techniques have also been applied to edge detection , e.g. , holistic edge detection [ 1 0 6 ] . | [
[
"edge detection",
"Task"
],
[
"holistic edge detection",
"Task"
]
] | [
[
"holistic edge detection",
"Used-For",
"edge detection"
]
] | ner: edge detection|Task ; holistic edge detection|Task || rel: holistic edge detection|Used-For|edge detection |
104291983 | The fully convolutional network is extended , by replacing a few ( typically two ) strided convolutions and the associated convolutions with dilated convolutions , to the dilation version , leading to medium - resolution representations [ 1 2 6 , 1 3 , 1 1 5 , 1 2 , 5 7 ] . | [
[
"fully convolutional network",
"Method"
],
[
"strided convolutions",
"Method"
],
[
"convolutions",
"Method"
],
[
"dilated convolutions",
"Method"
]
] | [] | ner: fully convolutional network|Method ; strided convolutions|Method ; convolutions|Method ; dilated convolutions|Method || rel: none |
104291983 | The upsample subnetwork could be a symmetric version of the downsample subnetwork , with skipping connection over some mirrored layers to transform the pooling indices , e.g. , SegNet [ 2 ] and DeconvNet [ 7 4 ] , or copying the feature maps , e.g. , U - Net [ 8 3 ] and Hourglass [ 7 2 , 1 1 1 , 7 , 2 2 , 6 ] , encoder... | [
[
"SegNet",
"Method"
],
[
"DeconvNet",
"Method"
],
[
"U - Net",
"Method"
],
[
"Hourglass",
"Method"
],
[
"encoder - decoder",
"Method"
],
[
"FPN",
"Method"
]
] | [] | ner: SegNet|Method ; DeconvNet|Method ; U - Net|Method ; Hourglass|Method ; encoder - decoder|Method ; FPN|Method || rel: none |
104291983 | The full - resolution residual network [ 7 8 ] introduces an extra full - resolution stream that carries information at the full image resolution , to replace the skip connections , and each unit in the downsample and upsample subnetworks receives information from and sends information to the full - resolution stream . | [
[
"full - resolution residual network",
"Method"
],
[
"full - resolution stream",
"Method"
],
[
"skip connections",
"Method"
],
[
"full - resolution stream",
"Method"
]
] | [
[
"full - resolution residual network",
"Part-Of",
"full - resolution residual network"
]
] | ner: full - resolution residual network|Method ; full - resolution stream|Method ; skip connections|Method ; full - resolution stream|Method || rel: full - resolution residual network|Part-Of|full - resolution residual network |
104291983 | Other works include : light upsample process [ 5 ] ; light downsample and heavy upsample processes [ 9 7 ] , recombinator networks [ 4 0 ] ; improving skip connections with more or complicated convolutional units [ 7 6 , 1 2 5 , 4 2 ] , as well as sending information from low - resolution skip connections to high - res... | [
[
"light upsample process",
"Method"
],
[
"light downsample and heavy upsample processes",
"Method"
],
[
"recombinator networks",
"Method"
],
[
"skip connections",
"Method"
],
[
"convolutional units",
"Method"
],
[
"low - resolution skip connections",
... | [] | ner: light upsample process|Method ; light downsample and heavy upsample processes|Method ; recombinator networks|Method ; skip connections|Method ; convolutional units|Method ; low - resolution skip connections|Method ; high - resolution skip connections|Method ; upsample process|Method ; multi - scale pyramid represe... |
104291983 | Representative works include GridNet [ 3 0 ] , convolutional neural fabrics [ 8 6 ] , interlinked CNNs [ 1 3 2 ] , and the recently - developed high - resolution networks ( HRNet ) [ 9 1 ] that is our interest . | [
[
"GridNet",
"Method"
],
[
"convolutional neural fabrics",
"Method"
],
[
"interlinked CNNs",
"Method"
],
[
"high - resolution networks",
"Method"
],
[
"HRNet",
"Method"
]
] | [
[
"HRNet",
"Synonym-Of",
"high - resolution networks"
]
] | ner: GridNet|Method ; convolutional neural fabrics|Method ; interlinked CNNs|Method ; high - resolution networks|Method ; HRNet|Method || rel: HRNet|Synonym-Of|high - resolution networks |
104291983 | The two early works , convolutional neural fabrics [ 8 6 ] and interlinked CNNs [ 1 3 2 ] , lack careful design on when to start low - resolution parallel streams and how and when to exchange information across parallel streams , and do not use batch normalization and residual connections , thus not showing satisfactor... | [
[
"convolutional neural fabrics",
"Method"
],
[
"interlinked CNNs",
"Method"
],
[
"batch normalization",
"Method"
],
[
"residual connections",
"Method"
]
] | [] | ner: convolutional neural fabrics|Method ; interlinked CNNs|Method ; batch normalization|Method ; residual connections|Method || rel: none |
104291983 | GridNet [ 3 0 ] is like a combination of multiple U - Nets and includes two symmetric information exchange stages : the first stage only passes information from high - resolution to low - resolution , and the second stage only passes information from low - resolution to high - resolution . | [
[
"GridNet",
"Method"
],
[
"U - Nets",
"Method"
]
] | [
[
"U - Nets",
"Part-Of",
"GridNet"
]
] | ner: GridNet|Method ; U - Nets|Method || rel: U - Nets|Part-Of|GridNet |
104291983 | The high - resolution network [ 9 1 ] , which we named HRNetV 1 for convenience , maintains high - resolution representations by connecting high - to - low resolution convolutions in parallel , where there are repeated multi - scale fusions across parallel convolutions . | [
[
"high - resolution network",
"Method"
],
[
"HRNetV 1",
"Method"
],
[
"high - to - low resolution convolutions",
"Method"
],
[
"parallel convolutions",
"Method"
]
] | [
[
"HRNetV 1",
"Part-Of",
"high - resolution network"
],
[
"high - to - low resolution convolutions",
"Part-Of",
"high - resolution network"
]
] | ner: high - resolution network|Method ; HRNetV 1|Method ; high - to - low resolution convolutions|Method ; parallel convolutions|Method || rel: HRNetV 1|Part-Of|high - resolution network ; high - to - low resolution convolutions|Part-Of|high - resolution network |
104291983 | The multi - resolution group convolution is a simple extension of the group convolution , which divides the input channels into several subsets of channels and performs a regular convolution over each subset over different spatial resolutions separately . | [
[
"multi - resolution group convolution",
"Method"
],
[
"group convolution",
"Method"
],
[
"convolution",
"Method"
]
] | [] | ner: multi - resolution group convolution|Method ; group convolution|Method ; convolution|Method || rel: none |
104291983 | A regular convolution can be divided as multiple small convolutions as explained in [ 1 2 2 ] . | [
[
"convolution",
"Method"
],
[
"small convolutions",
"Method"
]
] | [
[
"small convolutions",
"Part-Of",
"convolution"
]
] | ner: convolution|Method ; small convolutions|Method || rel: small convolutions|Part-Of|convolution |
104291983 | The differences lie in two - fold . ( i ) In a multi - resolution convolution each subset of channels is over a different resolution . ( ii ) The connection between input channels and output channels needs to handle The resolution decrease is implemented in [ 9 1 ] by using several 2 - strided 3 × 3 convolutions . | [
[
"multi - resolution convolution",
"Method"
],
[
"2 - strided 3 3 convolutions",
"Method"
]
] | [] | ner: multi - resolution convolution|Method ; 2 - strided 3 3 convolutions|Method || rel: none |
104291983 | In the original approach HRNetV 1 , only the representation ( feature maps ) from the high - resolution convolutions in [ 9 1 ] are outputted , which is illustrated in Figure 3 ( a ) . | [
[
"HRNetV 1",
"Method"
],
[
"high - resolution convolutions",
"Method"
]
] | [] | ner: HRNetV 1|Method ; high - resolution convolutions|Method || rel: none |
104291983 | In application to object detection , we construct a multi - level representation by downsampling the highresolution representation with average pooling to multiple levels , which is depicted in Figure 3 ( c ) . | [
[
"object detection",
"Task"
],
[
"average pooling",
"Method"
]
] | [
[
"average pooling",
"Used-For",
"object detection"
]
] | ner: object detection|Task ; average pooling|Method || rel: average pooling|Used-For|object detection |
104291983 | We name the two modifications as HRNetV 2 and HRNetV 2 p , respectively , and empirically compare them in Section 4. 4 . | [
[
"HRNetV 2",
"Method"
],
[
"HRNetV 2 p",
"Method"
]
] | [] | ner: HRNetV 2|Method ; HRNetV 2 p|Method || rel: none |
104291983 | In applications to semantic segmentation and facial landmark detection , we mix the output representations ( Figure 3 ( b ) ) , from all the four resolutions through a 1 × 1 convolution , and produce a 1 5 C - dimensional representation . | [
[
"semantic segmentation",
"Task"
],
[
"facial landmark detection",
"Task"
],
[
"1 1 convolution",
"Method"
]
] | [] | ner: semantic segmentation|Task ; facial landmark detection|Task ; 1 1 convolution|Method || rel: none |
104291983 | Then , we pass the mixed representation at each position to a linear classifier/regressor with the softmax/MSE loss to predict the segmentation maps/facial landmark heatmaps . | [
[
"linear classifier/regressor",
"Method"
],
[
"softmax/MSE loss",
"Method"
]
] | [
[
"softmax/MSE loss",
"Part-Of",
"linear classifier/regressor"
]
] | ner: linear classifier/regressor|Method ; softmax/MSE loss|Method || rel: softmax/MSE loss|Part-Of|linear classifier/regressor |
104291983 | For semantic segmentation , the segmentation maps are upsampled ( 4 times ) to the input size by bilinear upsampling for both training and testing . | [
[
"semantic segmentation",
"Task"
],
[
"segmentation",
"Task"
]
] | [
[
"segmentation",
"Used-For",
"semantic segmentation"
]
] | ner: semantic segmentation|Task ; segmentation|Task || rel: segmentation|Used-For|semantic segmentation |
104291983 | In application to object detection , we reduce the dimension of the high - resolution representation to 2 5 6 , similar to FPN [ 6 2 ] , through a 1 × 1 convolution before forming the feature pyramid in Figure 3 ( c ) . | [
[
"object detection",
"Task"
],
[
"FPN",
"Method"
],
[
"1 1 convolution",
"Method"
]
] | [] | ner: object detection|Task ; FPN|Method ; 1 1 convolution|Method || rel: none |
104291983 | We report the results over two scene parsing datasets , PASCAL Context [ 7 1 ] and Cityscapes [ 1 9 ] , and a human parsing dataset , LIP [ 3 4 ] . | [
[
"scene parsing",
"Task"
],
[
"PASCAL Context",
"Dataset"
],
[
"Cityscapes",
"Dataset"
],
[
"human parsing",
"Task"
],
[
"LIP",
"Dataset"
]
] | [
[
"PASCAL Context",
"Benchmark-For",
"scene parsing"
],
[
"Cityscapes",
"Benchmark-For",
"scene parsing"
],
[
"LIP",
"Benchmark-For",
"human parsing"
]
] | ner: scene parsing|Task ; PASCAL Context|Dataset ; Cityscapes|Dataset ; human parsing|Task ; LIP|Dataset || rel: PASCAL Context|Benchmark-For|scene parsing ; Cityscapes|Benchmark-For|scene parsing ; LIP|Benchmark-For|human parsing |
104291983 | The data are augmented by random cropping ( from 1 0 2 4 × 2 0 4 8 to 5 1 2 × 1 0 2 4 ) , random scaling in the range of [ 0 . 5 , 2 ] , and random horizontal flipping . | [
[
"random cropping",
"Method"
],
[
"random scaling",
"Method"
],
[
"random horizontal flipping",
"Method"
]
] | [] | ner: random cropping|Method ; random scaling|Method ; random horizontal flipping|Method || rel: none |
104291983 | We use the SGD optimizer with the base learning rate of 0.0 1 , the momentum of 0. 9 and the weight decay of 0.0 0 0 5 . | [
[
"SGD optimizer",
"Method"
],
[
"momentum",
"Method"
],
[
"weight decay",
"Method"
]
] | [] | ner: SGD optimizer|Method ; momentum|Method ; weight decay|Method || rel: none |
104291983 | Table 1 provides the comparison with several representative methods on the Cityscapes validation set in on ImageNet 3 and has similar model size as most DilatedResNet - 1 0 1 based methods . | [
[
"Cityscapes validation",
"Dataset"
],
[
"ImageNet 3",
"Dataset"
],
[
"DilatedResNet - 1 0 1",
"Method"
]
] | [] | ner: Cityscapes validation|Dataset ; ImageNet 3|Dataset ; DilatedResNet - 1 0 1|Method || rel: none |
104291983 | The data augmentation and learning rate policy are the same as Cityscapes . | [
[
"data augmentation",
"Method"
],
[
"Cityscapes",
"Dataset"
]
] | [] | ner: data augmentation|Method ; Cityscapes|Dataset || rel: none |
104291983 | In both cases , HRNetV 2 - W 4 8 performs su - 3 The description about ImageNet pretraining is given in the Appendix . [ 6 6 ] , the images are resized to 4 7 3 × 4 7 3 and the performance is evaluated on the average of the segmentation maps of the original and flipped images . | [
[
"HRNetV 2 - W 4 8",
"Method"
],
[
"ImageNet",
"Dataset"
]
] | [
[
"HRNetV 2 - W 4 8",
"Evaluated-With",
"ImageNet"
]
] | ner: HRNetV 2 - W 4 8|Method ; ImageNet|Dataset || rel: HRNetV 2 - W 4 8|Evaluated-With|ImageNet |
104291983 | The data augmentation and learning rate policy are the same as Cityscapes . | [
[
"data augmentation",
"Method"
],
[
"Cityscapes",
"Dataset"
]
] | [] | ner: data augmentation|Method ; Cityscapes|Dataset || rel: none |
104291983 | We apply our multi - level representations ( HRNetV 2 p ) 4 , shown in Figure 3 ( c ) , in the Faster R - CNN [ 8 2 ] and Mask R - CNN [ 3 8 ] frameworks . | [
[
"HRNetV 2 p",
"Method"
],
[
"Faster R - CNN",
"Method"
],
[
"Mask R - CNN",
"Method"
]
] | [
[
"HRNetV 2 p",
"Part-Of",
"Faster R - CNN"
],
[
"HRNetV 2 p",
"Part-Of",
"Mask R - CNN"
]
] | ner: HRNetV 2 p|Method ; Faster R - CNN|Method ; Mask R - CNN|Method || rel: HRNetV 2 p|Part-Of|Faster R - CNN ; HRNetV 2 p|Part-Of|Mask R - CNN |
104291983 | We perform the evaluation on the MS - COCO 2 0 1 7 detection dataset , which contains ∼ 1 1 8 k images for training , 5k for validation ( val ) and ∼ 2 0 k testing without provided annotations ( test - dev ) . | [
[
"MS - COCO 2 0 1 7 detection",
"Dataset"
]
] | [] | ner: MS - COCO 2 0 1 7 detection|Dataset || rel: none |
104291983 | We train the models for both our HRNetV 2 p and the ResNet on the public mmdetection platform [ 1 1 ] with the provided training setup , except that we use the learning rate schedule suggested in [ 3 7 ] for 2 × . | [
[
"HRNetV 2 p",
"Method"
],
[
"ResNet",
"Method"
]
] | [] | ner: HRNetV 2 p|Method ; ResNet|Method || rel: none |
104291983 | In the Faster R - CNN framework , our networks perform better than ResNets with similar parameter and computation complexity : HRNetV 2 p - W 3 2 vs. ResNet - 1 0 1 - FPN , HRNetV 2 p - W 4 0 vs. ResNet - 1 5 2 - FPN , HRNetV 2 p - W 4 8 vs. X - 1 0 1 - 6 4 × 4 d - FPN . | [
[
"Faster R - CNN",
"Method"
],
[
"ResNets",
"Method"
],
[
"HRNetV 2 p - W 3 2",
"Method"
],
[
"ResNet - 1 0 1 - FPN",
"Method"
],
[
"HRNetV 2 p - W 4 0",
"Method"
],
[
"ResNet - 1 5 2 - FPN",
"Method"
],
[
"HRNetV 2 p - W 4 8",
"Method... | [
[
"HRNetV 2 p - W 3 2",
"Compare-With",
"ResNet - 1 0 1 - FPN"
],
[
"HRNetV 2 p - W 4 0",
"Compare-With",
"ResNet - 1 5 2 - FPN"
],
[
"HRNetV 2 p - W 4 8",
"Compare-With",
"X - 1 0 1 - 6 4 4 d - FPN"
]
] | ner: Faster R - CNN|Method ; ResNets|Method ; HRNetV 2 p - W 3 2|Method ; ResNet - 1 0 1 - FPN|Method ; HRNetV 2 p - W 4 0|Method ; ResNet - 1 5 2 - FPN|Method ; HRNetV 2 p - W 4 8|Method ; X - 1 0 1 - 6 4 4 d - FPN|Method || rel: HRNetV 2 p - W 3 2|Compare-With|ResNet - 1 0 1 - FPN ; HRNetV 2 p - W 4 0|Compare-With|R... |
End of preview. Expand in Data Studio
SciER Baseline Predictions
Predictions from full fine-tuned google/flan-t5-small on all SciER LLM splits.
Model: sercetexam9/scier-flan-t5-small-baseline
Files:
train.jsonl,dev.jsonl,test.jsonl,test_ood.jsonlmetrics.json
- Downloads last month
- 59