Instance Segmentation has very limited resolution but can look great when what it is detected is small in screen size. This suggest for best performance a 8 Gb card with about 7 Gb of free memory will give best results with Rotobot. The amount of memory free will need to be 300Mb greater than the size of the model to allow for fluctuations in memory allocation. If you have 6.5 Gb of GPU memory free when the first node is computed you will have all nodes GPU accelerated. Memory allocation works as follows, once the first Rotobot node is computed, the Deep Neural network memory will be allocated and then recycled among the different models. When using the NVIDIA GPU for computation, the amount of memory required and allocated on the first use of any of the nodes will vary and it will use the most memory available on the card to account for as many models that may be used during a Rotobot session.įrom the the smallest model here are the memory requirements Rotobot Model When using a CPU only version of the software 16Gb of available random access memory should be sufficient. Rotobot ships with five deep learning models packaged in three plugins. Details of memory requirements for models ¶

Open fx plugins for davinci resolve