- implemented the papers model with help of AI. - Created 5k training scene (cornell box) with mitsuba renderer. - Tested object movements with a model trained with 400k iterations (5h training time with 4070ti), however results are not good, the light acts as if the objects are not moved from their place, like the model is memorized the static scene ![[Pasted image 20261018223239.png]] - After the initial cornell result, I tried training the same model with 5 different scenes, below is the result from testing that model against the single scene trained models. In dimining room scene, multi trained model carried too much of cornell boxes colors, I feel like this result comes from the similar cube architecture of the room. ![[dining-room.png]] ![[living-room-3.png]] ![[bathroom2.png]]