5.3. Results 5 69 Figure 5.4: Relative brightness of a phosphene in response to repeated stimulation, overlaid on experimental results by (Schmidt et al., 1996). The stimulation sequence consisted of 50 pulse trains at a four-second stimulation interval, followed by five pulse trains at an interval of 200 seconds to test recovery. The simulator was provided with a stimulation train of 125ms with a pulse width of 100µs at a frequency of 200Hz usinga stimulation amplitude of 90µA. Please notice the split x-axis with variable scaling. Figure 5.5: Performance as a function of resolution and number of phosphenes. The data is based on 5 runs of 1540 frames per condition, with batch size equal to 1 frame. Simulation was run with an NVIDIA© A30 GPU (memory size:24GB). Crosses indicate missing conditions. Note that these data are presented only for evaluating the software-performance. For some combinations of phosphene count and image resolution (e.g., 10.000 phosphenes in a 64×64 image) there are fewer pixels than phosphenes. in both hemifields, covering a field of view of 16 degrees of visual angle. Note that the simulated electrode density and placement differs from current prototype implants and the simulation can be considered to be an ambitious scenario from a surgical point of view, given the folding of the visual cortex and the part of the retinotopic map in V1 that is buried in the calcarine sulcus. We extended the previously published pipeline with 3D-convolutions (with an additional temporal dimension) to enable encoding of subsequent video frames. The model was trained on a basic video dataset with moving white digits on a black background (Moving MNIST Dataset; Srivastava et al., 2015).We used video sequences of five frames. The framerate of the simulation was set at five frames per second. We used a combination of two equally-weighted mean squared error (MSE) loss functions: the MSE loss between reconstruction and input, and the MSE loss between the simulated phosphene representation and the input. Figure5.7displays several frames after training for 45 epochs (for a total of 810,000 training examples). We can observe that
RkJQdWJsaXNoZXIy MTk4NDMw