search

meyerls / FruitNeRF

[IROS24] Offical Code for "FruitNeRF: A Unified Neural Radiance Field based Fruit Counting Framework" - Inegrated into Nerfstudio
https://meyerls.github.io/fruit_nerf
GNU General Public License v3.0
278 stars 36 forks source link

Issue with rendering output videos #15

Closed nikinezakati closed 1 month ago

nikinezakati commented 2 months ago

I'm trying to render the RGB, segmentation, and point cloud video outputs. When using ns-render, I encounter an error related to self.proposal_sampler = None in fruit_nerf.py. To address this, I set self.proposal_sampler = UniformSamplerWithNoise(num_samples=self.num_inference_samples, single_jitter=True) and self.num_inference_samples = int(200) However, this results in a CUDA out-of-memory error.

I’m running the code on two GPUs, each with 16GB of memory, and I can only manage to run the rendering with self.num_inference_samples set to 20 or lower. The output video doesn’t display any objects—it just shows a spectrum of colors. I'm testing on the synthetic dataset, specifically on the apple tree. The ns-viewer produced correct results during visualization in RGB format.

Could you please provide your visualization scripts or methods to resolve this issue?

meyerls commented 1 month ago

Apparently the extracted images were appended to a list and were not taken from the GPU. Thus the VRAM always exceeded with different types of settings.

Could you please try it again?