Hi @nickgillian,
i'm trying to run grt on esp-wroom-32 under free rtos.
I think i ported it successfully- project is building and working, I can load learned network from file and start prediction. I prepared DTW network, recognising two gestures with null rejection. In final version i want more gestures. As a time series i used data from 9 AOE sensor - linear acceleration for x,y,z axes and rotation quaternion, recorded during one sec on 100Hz. Its 7x100 array.
Main problem is lack of heap memory- i have 149184 bytes free before model loading, its enough for model and part of gesture, ~0,7 sec of record. Its too short for me.
Is possible to optimise library for RAM usage? Where I could start looking for it? What will happens if i will prepare model with more gestures? Its all loaded to RAM, or could i split it somehow and load needed part only?
Regards,
Grzegorz
Hi @nickgillian, i'm trying to run grt on esp-wroom-32 under free rtos. I think i ported it successfully- project is building and working, I can load learned network from file and start prediction. I prepared DTW network, recognising two gestures with null rejection. In final version i want more gestures. As a time series i used data from 9 AOE sensor - linear acceleration for x,y,z axes and rotation quaternion, recorded during one sec on 100Hz. Its 7x100 array. Main problem is lack of heap memory- i have 149184 bytes free before model loading, its enough for model and part of gesture, ~0,7 sec of record. Its too short for me. Is possible to optimise library for RAM usage? Where I could start looking for it? What will happens if i will prepare model with more gestures? Its all loaded to RAM, or could i split it somehow and load needed part only? Regards, Grzegorz