视频拍摄知识+AIGC数据预处理

视角

参考链接：https://www.polarpro.com/blogs/polarpro/filmmaking-101-types-of-camera-shots-and-angles

1. Low Angle Shot 低角度拍摄、
2. horizontal Shot 平视、
3. Dutch Angle Shot 荷兰角斜拍、
4. High Angle Shot 高角度拍摄、
5. Bird’s-eye / Aerial Shot 鸟瞰 / 航拍

运镜：

参考链接：https://www.storyblocks.com/resources/tutorials/7-basic-camera-movements

1. 变焦 Zoom （放大、缩小）
2. 平移 Pan （全景，左右旋转）
3. 倾斜 Tilt （上下旋转）
4. 小车 Dolly （相对主体前后平移）
5. 卡车 Truck （左右平移）
6. 基座 Pedestal （上下平移）
7. 焦点变换 Rack focus （画面大小不变，调整镜头以模糊一个主体，同时锐化另一个主体）
8. 固定 Fixed （镜头不动）

景别

参考链接：https://images.wikia.com/scratchpad/zh/images/f/f5/%E6%94%9D%E5%BD%B1%E9%8F%A1%E9%A0%AD%E4%B8%80%E8%A6%BD%E8%A1%A8.pdf

物体可分为：

1. 远景 Long Shot
2. 全景 Medium Long Shot、Panorama
3. 特写 Close-up
4. 微距 Macro shot

人物可细分：

1. 大远景 Extra Long Shot
2. 远景 Long Shot
3. 全景 Medium Long Shot、Panorama
4. 中景 Medium-shot
5. 近景 Medium Close-Up
6. 特写 Close-up
7. 微距 Macro shot

（就拿拍摄人物来说，远景就是把整个人和环境拍摄在画面里面，常用来展示事件发生的时间、环境、规模和气氛。往往是用来塑造一个大的场景。全景就是比远景更近一点，把人物的身体整个展示在画面里面，用来表现人物的全身动作，或者是人物之间的关系。中景就是指拍摄人物膝盖至头顶的部分，这样不仅能够展示人物的表情，而且还可以显示人物的形体动作。近景也就是拍摄人物胸部以上至头部的部位，非常有利于表现人物的面部或者是其他部门的表情、神态。包括眼角的细微动作。）

prompt test

使用视觉语言模型DeepseekVL-7B对视频的前中后三帧图像进行标注，然后使用大语言模型chatgpt对三个标注进行综合得到一段视频prompt。最后针对该prompt进行LLM问答，得到视频的拍摄信息。

DeepseekVL的prompt为：You are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible.

chatgpt的prompt为：You are a powerful video captioner. The next 3 phases describe the first, middle, and end frame of the video. Please summarize the following 3 paragraphs into 1 paragraph, which can provide a detailed description of the video information.

1. 视角

prompt

there is a caption of a video: The video begins with an aerial view of a vibrant suburban neighborhood, showcasing approximately 50 houses, 10 multi-story apartment buildings, along with distinct red-roofed and green-roofed structures amidst lush greenery and tree-lined streets. Transitioning to a historical fortification in the middle frame, the scene unveils a grand stone wall encircling a spacious green area, punctuated by scattered trees and a central prominent tree, juxtaposed against modern buildings in the backdrop. Finally, the video concludes with another aerial perspective, this time capturing a historic fortification nestled within a hilly landscape, featuring stone buildings with thatched roofs, picnic areas amidst a grassy terrain, and distant houses and trees, all illustrating the captivating blend of history and nature. Each phase offers a unique glimpse into different environments, from suburban living to historical landmarks, all captured through stunning aerial photography.

The shooting angles of a video are divided into: