Confusing warning in "rasterize_coarse.cu"

[Angtian]

The warning in rasterize_coarse.cu. Line 193 is confusing. It says:
"Bin size was too small in the coarse rasterization phase. "
"This caused an overflow, meaning output may be incomplete. "
"To solve, "
"try increasing max_faces_per_bin / max_points_per_bin, "
"decreasing bin_size, "
"or setting bin_size to -1 to use the naive rasterization.";
It is my understanding that it warning appears when Bin size is large than excepted, not small.

[gkioxari]

Our rasterizer takes a coarse-to-fine approach by binning the image into rectangles to compute the rasterization outputs.
We set the bin size by default but if your mesh/point cloud is too large then the default value might not be appropriate.
This warning tells you exactly that. So you need to increase the max_faces_per_bin (for meshes) or the max_points_per_bin (for point clouds).

[Angtian]

Thanks for the reply. I get the point. But may I have one more question, In the first sentence it says "Bin size was too small", but it also suggests "decreasing bin_size", I am confused about this since it seems to be a conflict.

[gkioxari]

There are two parameters here: bin_size and max_faces_per_bin.
If your mesh is too large for our default values, you can either increase max_faces_per_bin or decrease bin_size or as the comment says set bin_size = -1 to get the naive rasterization which does not follow a coarse-to-fine strategy but is slower.

You can find their definitions here:

pytorch3d/pytorch3d/renderer/mesh/rasterizer.py

Lines 37 to 45 in c2862ff

	bin_size: Size of bins to use for coarse-to-fine rasterization. Setting
	bin_size=0 uses naive rasterization; setting bin_size=None attempts
	to set it heuristically based on the shape of the input. This should
	not affect the output, but can affect the speed of the forward pass.
	max_faces_per_bin: Only applicable when using coarse-to-fine
	rasterization (bin_size != 0); this is the maximum number of faces
	allowed within each bin. This should not affect the output values,
	but can affect the memory usage in the forward pass.
	Setting max_faces_per_bin=None attempts to set with a heuristic.

[Angtian]

Then for your reply. So if my understanding is correct that this warning means the "bin size" is set to be too large that we need to decrease it?

[gkioxari]

Yes. Or increase max_faces_per_bin.

[JLee2022]

excellent！help me a lot

[Veni-VidiVici-xiaoxu]

so what number shall I take for bin_size and max_points_per_bin?

[HernandoR]

so what number shall I take for bin_size and bin_size?

furtheraway, what whould be the difference between increasing bin_size and bin_size?

[wjh19990923]

but making bin_size=0 will significantly increase the computational time, is there any way to switch off the warning because its annoying in iterative optimization

[HernandoR]

one way (perhaps the only way) of doing so is to recompile by yourself. I’ve done this once since could not find appropriate version for pip management. the warning is just few lines in the cuda source file. Liu Zhen

…

________________________________ From: Jinhao Wang ***@***.***> Sent: Friday, December 6, 2024 9:28:03 PM To: facebookresearch/pytorch3d ***@***.***> Cc: Liu Zhen ***@***.***>; Comment ***@***.***> Subject: Re: [facebookresearch/pytorch3d] Confusing warning in "rasterize_coarse.cu" (Issue #1064) but making bin_size=0 will significantly increase the computational time, is there any way to switch off the warning because its annoying in iterative optimization — Reply to this email directly, view it on GitHub<#1064 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AK4XSWB5QKVZNC5JR5PTN2T2EGQ6HAVCNFSM6AAAAABMCMWJBWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMRTGI2TOMZSGI>. You are receiving this because you commented.Message ID: ***@***.***>