TensorFlow.js Model Benchmark
{ "core": "4.22.0", "layers": "4.22.0", "converter": "4.22.0" }
Parameters
Type
Value
Inputs
Type
Value
Model
Type
Value
Inference times
Kernel
Type
Time(ms)
Controls
Benchmark
models
MobileNetV3
MobileNetV2
MobileNetV2Lite
HandPoseDetector
HandPoseLandmark
MoveNet-SinglePose
MoveNet-MultiPose
BlazePoseDetector
BlazePoseLandmark
Coco-SSD
DeepLabV3
FaceDetection
FaceLandmarkDetection
ArPortraitDepth
SelfieSegmentation-General
SelfieSegmentation-Landscape
AutoML Image
AutoML Object
USE - batchsize 30
USE - batchsize 1
TextToxicity
MobileBert
posenet
bodypix
speech-commands
custom
MobileNetV3
Parameters
numWarmups
numRuns
numProfiles
kernelTiming
aggregate
individual
aggregate
parallelCompile
Model Parameters
architecture
small_075
small_100
large_075
large_100
small_075
Environment
backend
wasm
webgl
cpu
webgpu
tflite
webgl
cpu forward
enforce float16
GL flush wait time(ms)
-1
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
-1
Packed depthwise Conv2d
Use shapes uniforms
Print intermediate tensors
Run benchmark
Test correctness