Bug: Requested texture size greater than WebGL maximum on this browser / GPU.

[dipterix]

I was trying to run on my own data when I got this error:

I'm using Apple M2 Air, Firefox.

[sergeyplis]

Hi @dipterix which model are you running? Can you post Diagnostics here? Click on the diagnostics button after you got this error and it will be copied into your buffer as below:

:: Diagnostics can help resolve issues https://github.com/neuroneural/brainchop/issues ::
TF_Backend: webgl
isModelFullVol: true
No_SubVolumes: 1
Brainchop_Ver: FullVolume
Input_Shape: [null,200,200,260,1]
Output_Shape: [null,256,256,256,3]
Channel_Last: true
Model_Param: 5598
Model_Layers: 20
Model: ⚡ Tissue GWM (light)
Actual_Labels: 3
Expect_Labels: 3
NumLabels_Match: true
Inference_t: 0.3730
Postprocess_t: 0.4700
Status: OK
Total_t: 1.21
Date: 9/29/2024
Browser: Chrome
Browser_Ver: 128
OS: MacOS
WebGL2: true
GPU_Vendor: Apple
GPU_Card: ANGLE Metal Renderer: Apple M2 Max
GPU_Vendor_Full: Google Inc. (Apple)
GPU_Card_Full: ANGLE (Apple, ANGLE Metal Renderer: Apple M2 Max, Unspecified Version)
CPU_Cores: 5
Which_Brainchop: latest
Heap_Size_MB: 64.24498844146729
Used_Heap_MB: 61.1394624710083
Heap_Limit_MB: 4095.75
Texture_Size: 16384

[dipterix]

Sure!

Here are results using firefox and chrome:

Diagnostics copied to clipboard
:: Diagnostics can help resolve issues https://github.com/neuroneural/brainchop/issues ::
TF_Backend: webgl
isModelFullVol: true
No_SubVolumes: 1
Brainchop_Ver: FullVolume
Input_Shape: [null,288,241,288,1]
Output_Shape: [null,256,256,256,3]
Channel_Last: true
Model_Param: 5598
Model_Layers: 20
Model: ⚡ Brain Mask (FAST)
Extra_Info: null
Inference_t: Infinity
Postprocess_t: Infinity
Status: Fail
Error_Type: Requested texture size [9997x9998] greater than WebGL maximum on this browser / GPU [8192x8192].
Extra_Err_Info: Failed while model layer 1 apply
Total_t: 7.37
Date: 10/1/2024
Browser: Firefox
Browser_Ver: 130
OS: MacOS
WebGL2: true
GPU_Vendor: Apple
GPU_Card: Apple M1, or similar
GPU_Vendor_Full: Apple
GPU_Card_Full: Apple M1, or similar
CPU_Cores: 8
Which_Brainchop: latest
Texture_Size: 8192

I switched to Chrome and the model passed fine. However the result is not right:

[sergeyplis]

Great! Note, the models marked with ⚡ symbol are the smallest fastest and thus less robust :)
Try models with 🪓 - they are much more robust. Also, note that all models have different modes of running. If the High Memory Fast version fails - try Low Memory Slower version. It is the same model and you will get exactly the same result. The only difference is how the model is executed.

[dipterix]

The issue is none of the models worked. All models can only detect very small brain areas.

[sergeyplis]

Interesting. Is this a publicly available sample or can you share it otherwise?