
Grid Image

Rays Image

Sample Points Image

Part 1 gif

Depth Image

Part 2 gif

Part 3 gif

Part 4.1 gif
In reality the color can change a little bit because of the lighting when the viewpoint changes. Adding viewpoint dependence can make this happen. If the view dependence is too strong, the network can memorize the training cameras instead of learning a color that stays consistent in 3D. If it is too weak, the model generalizes more easily, but the reflection may be less sharp.

Part 5 Render gif
{'params': [gaussians.pre_act_opacities], 'lr': 0.05, "name": "opacities"}, {'params': [gaussians.pre_act_scales], 'lr': 0.005, "name": "scales"}, {'params': [gaussians.colours], 'lr': 0.01, "name": "colours"}, {'params': [gaussians.means], 'lr': 0.0001, "name": "means"}
1000 iterations
Mean PSNR: 29.398, mean SSIM: 0.936.

Part 6 Training Progress gif

Part 6 Test Results gif

Part 5 Render gif

Part 7.1 Render gif

side-by-sdie 1

side-by-sdie 2

side-by-sdie 3
1: the upper area is different, sh one is darker. 2: the upper area is different, sh one is darker but slightly lighter than the previous one. 3: the upper image is different. The sh one is darker but lighter than the previous two. The original one does not change when the viewpoint changes, but the sh one can change.