List your company
ResearchFormerly WizWand

Image Classification on Birdsnap (test)

85.5Top-1 AccJFT - Bird
35.47648.46361.4574.437Nov 16, 2018May 27, 2019Dec 5, 2019Jun 15, 2020Dec 24, 2020Jul 4, 2021Jan 13, 2022
Updated 8mo ago

Evaluation results

MethodLinks
JFT - BirdBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1185.5
Gradient-based FVEpart detector=CS-Parts [24], optimization=gradient descent2020.0785.3
JFT - Adaptive TransferBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT2018.1185.1
EM-based FVEpart detector=CS-Parts [24], optimization=online EM algorithm2020.0784.9
FixSENet-1542020.0784.3
JFT - AnimalBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1184.1
GAPpart detector=CS-Parts [24], aggregation=concatenated part features and GAP2020.0784
Best Published Result [17]Protocol=Reference value2018.1183.9
GPipe (AmoebaNet-B (18, 512))Architecture=AmoebaNet-B (18, 512), Resolution=480x480, Protocol=Fine-tuned, Number of fine-tuning runs=5, Crop=Single-crop2018.1183.6
No Parts (baseline)parts=none, features=global image only2020.0781.9
ImageNet - Entire DatasetBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=ImageNet2018.1180.8
ImageNet - Adaptive TransferBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=ImageNet2018.1180.7
Entire JFT DatasetBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT2018.1180.3
Method [32]2018.1180.2
JFT - FoodBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1179.7
JFT - TransportBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1179.2
JFT - CarBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1179
JFT - VehicleBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1178.8
JFT - AircraftBackbone=AmoebaNet-B, Input Resolution=331 x 331, Pre-training Source=JFT (Hand-selected subset)2018.1178
ReLICv2Evaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0176.7
BYOLEvaluation Protocol=Fine-tuned2021.0676.3
BYOLEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0176.3
Random initEvaluation Protocol=Fine-tuned2021.0676.1
Random InitEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=None2022.0176.1
SimCLREvaluation Protocol=Fine-tuned2021.0675.9
SimCLREvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0175.9
Supervised-INEvaluation Protocol=Fine-tuned2021.0675.8
Supervised-INEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0175.8
SSL-HSIC (w/ target)Evaluation Protocol=Fine-tuned2021.0674.9
SSL-HSIC (w/o target)Evaluation Protocol=Fine-tuned2021.0673.1
ReLICv2Evaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0165.4
NNCLREvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0161.4
SSL-HSIC (w/ target)Evaluation Protocol=Linear2021.0657.8
BYOLEvaluation Protocol=Linear2021.0657.2
BYOLEvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0157.2
Supervised-INEvaluation Protocol=Linear2021.0653.7
Supervised-INEvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0153.7
SSL-HSIC (w/o target)Evaluation Protocol=Linear2021.0650.6
ReLICv2Backbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=5000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0149.4
ReLICv2Backbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=1000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0147.4
DnCBackbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=4500 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0142.1
BYOLBackbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=5000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0138.2
SimCLREvaluation Protocol=Linear2021.0637.4
SimCLREvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0137.4