The main focus for this week was correcting the errors in the bounding box drawn. The bounding boxes differed from the ground truth. That was because of issue regarding the canvas context drawing setting and not related to the actual model.
After the correcting the drawing steps in the canvas context. The bounding box are drawn somewhat similar to the ground truth.
Second goal for this week was development of a remote server based object detection video api. I successfully mad it work for single image but beyond it the server started to hang-up.
The video encoding and decoding process are particularly stressful for the CPU. Throughout the development time the CPU was under 96~97% utilisation running at maximum clock speed. This was the case for a single instance, but if multiple instances are launched the server crashes instantaneously. For running the frozen inference model GPU was used. Under single gpu setup the gpu was under 100% memory utilisation.
Apart from this network latency spikes up during the encoding and decoding process.
That's all for today!!!!
Hope you had a great week