MogaNet: Multi-order Gated Aggregation Network

Open paper
Benchmark Model Rank Results
video-prediction-on-moving-mnistMogaNet (SimVP 10x)#5MSE: 15.67MAE: 51.84SSIM: 0.9661
video-prediction-on-moving-mnistVAN (SimVP 10x)#7MSE: 16.21MAE: 53.57SSIM: 0.9646
video-prediction-on-moving-mnistHorNet (SimVP 10x)#8MSE: 17.4MAE: 55.7SSIM: 0.9624
video-prediction-on-moving-mnistConvNeXt (SimVP 10x)#9MSE: 17.58MAE: 55.76SSIM: 0.9617
video-prediction-on-moving-mnistUniformer (SimVP 10x)#11MSE: 18.01MAE: 57.52
video-prediction-on-moving-mnistMLP-Mixer (SimVP 10x)#13MSE: 18.85MAE: 59.86
video-prediction-on-moving-mnistSwin (SimVP 10x)#15MSE: 19.11MAE: 59.84
video-prediction-on-moving-mnistViT (SimVP 10x)#16MSE: 19.74MAE: 61.65SSIM: 0.9539
video-prediction-on-moving-mnistPoolformer (SimVP 10x)#18MSE: 20.96MAE: 64.31
video-prediction-on-moving-mnistConvMixer (SimVP 10x)#20MSE: 22.3MAE: 67.37
instance-segmentation-on-coco-test-devMogaNet-XL (Cascade Mask R-CNN)#27mask AP: 48.8
instance-segmentation-on-coco-test-devMogaNet-L (Cascade Mask R-CNN)#39mask AP: 46.1
instance-segmentation-on-coco-test-devMogaNet-B (Cascade Mask R-CNN)#40mask AP: 46
instance-segmentation-on-coco-test-devMogaNet-S (Cascade Mask R-CNN)#43mask AP: 45.1
instance-segmentation-on-coco-test-devMogaNet-L (Mask R-CNN 1x)#45mask AP: 44.1
instance-segmentation-on-coco-test-devMogaNet-B (Mask R-CNN 1x)#48mask AP: 43.2
instance-segmentation-on-coco-test-devMogaNet-S (Mask R-CNN 1x)#52mask AP: 42.2
instance-segmentation-on-coco-test-devMogaNet-T (Mask R-CNN 1x)#80mask AP: 39.1
instance-segmentation-on-coco-test-devMogaNet-XT#88mask AP: 37.6
instance-segmentation-on-coco-test-devMogaNet-T#94mask AP: 35.8
human-pose-keypoint-estimation-on-coco-val2017MogaNet-B (384x288)#33AP: 77.3AP50: 91.4AP75: 84.0AR: 82.2
human-pose-keypoint-estimation-on-coco-val2017MogaNet-S (384x288)#41AP: 76.4AP50: 91.0AP75: 83.3AR: 81.4
human-pose-keypoint-estimation-on-coco-val2017MogaNet-S (256x192)#59AP: 74.9AR: 80.1
human-pose-keypoint-estimation-on-coco-val2017MogaNet-T (256x192)#71AP: 73.2AP50: 90.1AP75: 81.0AR: 78.8
semantic-segmentation-on-ade20kMogaNet-XL (UperNet)#73Validation mIoU: 54
semantic-segmentation-on-ade20kMogaNet-L (UperNet)#117Validation mIoU: 50.9GFLOPs (512 x 512): 1176
semantic-segmentation-on-ade20kMogaNet-B (UperNet)#130Validation mIoU: 50.1GFLOPs (512 x 512): 1050
semantic-segmentation-on-ade20kMogaNet-S (UperNet)#148Validation mIoU: 49.2GFLOPs (512 x 512): 946
semantic-segmentation-on-ade20kMogaNet-S (Semantic FPN)#175Validation mIoU: 47.7GFLOPs (512 x 512): 189
image-classification-on-imagenetMogaNet-XL (384res)#78Top 1 Accuracy: 87.8%Number of params: 181MGFLOPs: 102
image-classification-on-imagenetMogaNet-L#320Top 1 Accuracy: 84.7%Number of params: 83MGFLOPs: 15.9
image-classification-on-imagenetMogaNet-B#356Top 1 Accuracy: 84.3%Number of params: 44MGFLOPs: 9.9
image-classification-on-imagenetMogaNet-S#464Top 1 Accuracy: 83.4%Number of params: 25MGFLOPs: 5
image-classification-on-imagenetMogaNet-T (256res)#764Top 1 Accuracy: 80%Number of params: 5.2MGFLOPs: 1.44
image-classification-on-imagenetMogaNet-XT (256res)#925Top 1 Accuracy: 77.2%Number of params: 3MGFLOPs: 1.04
object-detection-on-coco-val2017MogaNet-XL (IN-22K, Cascade Mask R-CNN 3x)#86box AP: 56.2AP50: 75.0AP75: 61.2Params (M): 238
object-detection-on-coco-val2017MogaNet-L (Cascade Mask R-CNN 3x)#145box AP: 53.3AP50: 71.8AP75: 57.8Params (M): 140
object-detection-on-coco-val2017MogaNet-B (Cascade Mask R-CNN 3x)#164box AP: 52.6AP50: 72.0AP75: 57.3Params (M): 101
object-detection-on-coco-val2017MogaNet-S (Cascade Mask R-CNN 3x)#183box AP: 51.6AP50: 70.8AP75: 56.3Params (M): 83
object-detection-on-coco-val2017MogaNet-L (Mask R-CNN 1x)#231box AP: 49.4AP50: 70.7AP75: 54.1Params (M): 102
object-detection-on-coco-val2017MogaNet-L (RetinaNet 1x)#248box AP: 48.7AP50: 69.5AP75: 52.6APS: 31.5APM: 53.4APL: 62.7…
object-detection-on-coco-val2017MogaNet-B (Mask R-CNN 1x)#264box AP: 47.9AP50: 70.0AP75: 52.7Params (M): 63
object-detection-on-coco-val2017MogaNet-B (RetinaNet 1x)#268box AP: 47.7AP50: 68.9AP75: 51.0APS: 30.5APM: 52.2APL: 61.7…
object-detection-on-coco-val2017MogaNet-S (Mask R-CNN 1x)#285box AP: 46.7AP50: 68.0AP75: 51.3Params (M): 45
object-detection-on-coco-val2017MogaNet-S (RetinaNet 1x)#298box AP: 45.8AP50: 66.6AP75: 49.0APS: 29.1APM: 50.1APL: 59.8…
object-detection-on-coco-val2017MogaNet-T (Mask R-CNN 1x)#359box AP: 42.6AP50: 64.0AP75: 46.4Params (M): 25
object-detection-on-coco-val2017MogaNet-T (RetinaNet 1x)#373box AP: 41.4AP50: 61.5AP75: 44.4APS: 25.1APM: 45.7APL: 53.6…
object-detection-on-coco-val2017MogaNet-XT (Mask R-CNN 1x)#391box AP: 40.7AP50: 62.3AP75: 44.4Params (M): 23
object-detection-on-coco-val2017MogaNet-XT (RetinaNet 1x)#406box AP: 39.7AP50: 60.0AP75: 42.4APS: 23.8APM: 43.6APL: 51.7…

Rank counts only results with a code link.