search

NVIDIA-AI-IOT / face-mask-detection

Face Mask Detection using NVIDIA Transfer Learning Toolkit (TLT) and DeepStream for COVID-19
MIT License
241 stars 94 forks source link

Low FPS in Jetson Nano (11 FPS) #17

Closed lakshaychhabra closed 3 years ago

lakshaychhabra commented 3 years ago

Hi, I trained the model and pruned it 30% (instead of 12% as mentioned in Readme) in hope of getting more FPS. But unfortunately getting only ~11FPS (FP16). Is there any chance of increasing FPS by pruning it to 80%?

Also, I am unable to detect mask as Bounding boxes are not displayed in Deepstream.

lakshaychhabra commented 3 years ago

Well, I changed the input size and FPS increased to 25.

craston commented 3 years ago

@lakshaychhabra Did you manage to detect the mask as bbox in Deepstream. Currently I do not see any bbox in Deepstream

lakshaychhabra commented 3 years ago

@craston yes I managed to detect the mask as bbox, [property] gpu-id=0 net-scale-factor=0.0039215697906911373 tlt-model-key="my-key" tlt-encoded-model=final_model.etlt labelfile-path=labels_masknet.txt

model-engine-file=final_model.etlt_b6_gpu0_fp16.engine input-dims=3;300;300;0 uff-input-blob-name=input_1 force-implicit-batch-dim=1 batch-size=1 network-mode=2 process-mode=1 num-detected-classes=2 interval=4 gie-unique-id=2 output-blob-names=output_bbox/BiasAdd;output_cov/Sigmoid cluster-mode=1

[class-attrs-all] pre-cluster-threshold=0.2 eps=0.3 minBoxes=1

can be tweaked

group-threshold=1

[application] enable-perf-measurement=1 perf-measurement-interval-sec=1

This was my updated config file, try changing [class-attrs-all] in your config file

craston commented 3 years ago

Thanks I got it to work!