Open naturezhanghn opened 4 months ago
In general scenarios where data comes in good quality, the VFI method combined with events shows significantly superior performance. However, the ERDS dataset assumes extreme conditions where event data may be placed in such situations. Therefore, evaluating including a general scene shows that event-based VFI performs better. Conversely, in situations where these conditions are continually changing events, event-based VFI may show poorer performance. We have attempted to handle this issue as effectively as possible using TTA techniques.
Thank you for your response. We greatly appreciate your contributions to the data and look forward to the code release.
Hello, I have been testing interpolation results on ERDS dataset and noticed that methods incorporating event data significantly underperform compared to pure RGB interpolation methods like RIFE and Super SloMo. Could you please explain why this might be the case? Is it possible that pure RGB interpolation is more effective for the specific data? Thank you for your insights.