Old Wardor Castle, Tisbury

This one was another experiment in extracting models from youtube drone footage, extracting frames on the half-second. As with the [windmill scan](https://sketchfab.com/3d-models/halnaker-windmill-sussex-3d30e185f0604a6f96f6f55d9572d07a), these ones are really hit and miss. It takes a lot of searching to find videos that have the ideal qualities for model generation - 4k video, arcing shots from every angle, slow enough speed to avoid blur, overcast day to avoid hard shadows, etc. Even some that come pretty close to ideal can still have missing areas. Theres a YT extracted scan of Osaka Castle I'm working on right now that was almost perfect.. but somehow can't fully resolve one of the sides. Surprisingly, watermarks and black bars don't cause major problems, though crossfades and other transitions can produce some bad frames. [Source Video](https://www.youtube.com/watch?v=1pAxnyKfBJo) by Stonehenge Dronescapes. Source: Objaverse 1.0 / Sketchfab本实验为一项基于YouTube无人机航拍影像的模型提取实验，采用每半秒提取一帧的采样方式。与此前的[风车扫描模型（windmill scan）](https://sketchfab.com/3d-models/halnaker-windmill-sussex-3d30e185f0604a6f96f6f55d9572d07a)项目一致，此类模型生成结果往往成败参半。 想要获取适配模型生成的优质视频素材需耗费大量搜寻精力：需满足4K分辨率、全角度环绕运镜、足够缓慢的拍摄速度以避免画面模糊、阴天拍摄以消除硬阴影等要求。即便部分视频近乎达到理想标准，仍可能存在区域缺失的问题。目前我正在处理一段从YouTube提取的大阪城扫描素材，其整体质量近乎完美，但不知为何始终无法完整还原其中一侧的结构。令人意外的是，水印与黑边并不会造成严重干扰，但交叉淡入淡出及其他转场效果可能会生成不合格帧。 [源视频（Source Video）](https://www.youtube.com/watch?v=1pAxnyKfBJo)由Stonehenge Dronescapes制作发布。 数据集来源：Objaverse 1.0 / Sketchfab