Is there any recommended way to speed up detection and recognition process for a large number of images?
Right now it's taking ~1.1 sec to process (detect & recognize) one 800x600 jpeg image on V100 (Amz deep learning ami).
I'm using a pipeline from jupyter notebook.
Using a batch of 4 images does not give any speedup compared to 1 image. Is that right?
When a pipeline is fresh loaded, VRAM consumption is 2Gb from 16Gb.
After I start processing, it jumps to 15.9Gb.
GPU is clearly underloaded, most of time nvidia-smi reporting up to 19% GPS usage, often 0-2%, sometimes 100%. CPU is moderately loaded all the time.
Thank you very much for this awesome package!
Is there any recommended way to speed up detection and recognition process for a large number of images? Right now it's taking ~1.1 sec to process (detect & recognize) one 800x600 jpeg image on V100 (Amz deep learning ami). I'm using a pipeline from jupyter notebook. Using a batch of 4 images does not give any speedup compared to 1 image. Is that right? When a pipeline is fresh loaded, VRAM consumption is 2Gb from 16Gb. After I start processing, it jumps to 15.9Gb. GPU is clearly underloaded, most of time nvidia-smi reporting up to 19% GPS usage, often 0-2%, sometimes 100%. CPU is moderately loaded all the time.