Open asedfrgt opened 2 years ago
Hi,I use mobile phone to capture a indoor scene, and walk slowly ahead with my phone. The phone is alway facing forward, tooks a full circle path indoors. I use the ns-process-data to process this data. After processed, the dateset has 1020 frames. And then, I use ns-train nerfacto --data {PROCESSED_DATA_DIR} to train the data, but the reslut of render image is blur with losting some details. testing psnr(20db) is lower than training psnr(28db).
Can you provide example example renders. Also a screenshot of the camera distribution (from the visualization in the viewer).
Thanks for your reply. Sorry I have not visualize the camera distribution in the viewer. The camera distribution is as shown in the following figure. The white indicates the camera distribution, and the red indicates the camera orientation. The render trajectory is the same as camera distribution.
If you are able to recapture the scene, you should try rotating as you walk through the scene in increase the number of angles that you view the scene.
I can get good result on this data using instant-ngp. Is the nerfacto method not suitable this data?
If it works works with NGP it should also do reasonably well with nerfacto. Can you share an example of your results to make debugging easier?
Hi, I use the nerfato method on desoltation dataset after 10w interations. The traning psnr is 25db, and the testing psnr is 20db. The result lost some detail and is blur. How can I get good result using nerfato method?