MrNeRF / awesome-3D-gaussian-splatting

Curated list of papers and resources focused on 3D Gaussian Splatting, intended to keep pace with the anticipated surge of research in the coming months.
MIT License
4.97k stars 279 forks source link

(not issue, sharing) generalized visual quality optimzation (updating ...) #118

Open yuedajiong opened 1 month ago

yuedajiong commented 1 month ago

updating ... (I will close or delete this issue once I stop updating) Your feedback is welcomed as I acknowledge my limitations.


Current topic, 'generalized visual quality':

  1. Once you have identified the main challenges of your paper (such as single-view/image, camera-pose-free, dynamic, ...), or once you intend to offer the reconstructed/generated output as a commercial product, you will immediately encounter the primary issue of "generalized visual quality." (SO FAR, NO PAPER CAN DO. JUST TOY OR BIGGER TOY.)
  2. Therefore, I am leveraging my limited understanding to attempt an analysis of all the technical aspects related to "generalized visual quality" and gathering relevant papers.

https://github.com/yuedajiong/super-ai/blob/main/superai-20240428-superv-visual-quality.txt


Next topic, I will focus on 'fucking limited resource':

  1. single beggar's GPU for train and infer (dataset: objvase-level 3d objects + internet-level 2d images)
  2. diff hash based priori (incremental)
  3. diff compressed GS respresentation
  4. pre-computed image features (eg. features on rendered objvase )
  5. ...

https://github.com/yuedajiong/super-ai/blob/main/superai-20240428-superv-limited-resource.txt