Android GPU Delegate w/ Selfie Multiclass gives "Error Batch size mismatch, expected 1 but got 8"

[peymannaderi10]

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

No

OS Platform and Distribution

Android 14

Mobile device if the issue happens on mobile device

Galaxy S21+ 5G, Model: SM-G996U/DS

Browser and version if the issue happens on browser

No response

Programming Language and version

Java/Kotlin

MediaPipe version

0.10.14

Bazel version

No response

Solution

Image Segmentation, Selfie Multiclass Model

Android Studio, NDK, SDK versions (if issue is related to building in Android environment)

Android Studio Meerkat | 2024.3.1 Patch 2, Runtime version: 21.0.6+-13355223-b631.42 amd64

Xcode & Tulsi version (if issue is related to building for iOS)

No response

Describe the actual behavior

Then upon opening the app, I would swipe up to select the ML model and delegate, selected Selfie Multiclass ML model, and then I would select GPU for delegate and on the screen I see "Image segmenter failed to initialize. See logs for error details", when going to logcat below is what would print: 2025-05-29 11:59:54.668 24517-24625 native com...pe.examples.imagesegmentation E E0000 00:00:1748534394.668891 24625 calculator_graph.cc:887] INVALID_ARGUMENT: CalculatorGraph::Run() failed: Calculator::Open() for node "mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: Batch size mismatch, expected 1 but got 8

Describe the expected behaviour

For the Selfie Multiclass model to work on android with GPU delegation

Standalone code/steps you may have used to try to get what you need

I downloaded the image segmentation category mask example and followed the README to install the demo app, filtered logcat to view the error.

Other info / Complete Logs

2025-05-29 11:59:54.668 24517-24625 native com...pe.examples.imagesegmentation E E0000 00:00:1748534394.668891 24625 calculator_graph.cc:887] INVALID_ARGUMENT: CalculatorGraph::Run() failed:
Calculator::Open() for node "mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: Batch size mismatch, expected 1 but got 8

2025-05-29 11:59:54.669 24517-24625 ImageSegmenterHelper com...pe.examples.imagesegmentation E Image segmenter failed to load model with error: invalid argument: CalculatorGraph::Run() failed:
Calculator::Open() for node "mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: Batch size mismatch, expected 1 but got 8

[kuaashish]

Hi @peymannaderi10,

Can you please let us know the complete steps you are following Or you can point out to documentation? This info will help us to understand the issue and reproduce it if required.

Thank you!!

[peymannaderi10]

@kuaashish yea of course! I will point out exactly what I did in the steps below:

Step 1. Install prerequisites such as Java Development Kit (JDK), Android Studio (version 2021.1.1 or later. Also have a physical Android device with a minimum OS version of SDK 24 (Android 7.0 - Nougat) with developer mode enabled.

Step 2. Clone the media-pipe samples repository and cd to the category mask for android image segmentation example:

git clone https://github.com/google-ai-edge/mediapipe-samples.git
cd mediapipe-samples/examples/image_segmentation/android/category_mask

Step 3. Open Android Studio. From the Welcome screen, select Open an existing Android Studio project. From here select the category mask folder in mediapipe-samples/examples/image_segmentation/android/category_mask, Click OK. You may be asked if you trust the project. Select Trust.

Step 4. If it asks you to do a Gradle Sync, click OK. If it does not you can press file at the top left and then press on "Sync Project with Gradle Files", or press Ctrl+Shift+O.

Step 5. With your Android device connected to your computer and developer mode enabled, click on the green Run arrow in Android Studio or press Shift + F10.

Step 6. Wait for the app to build and open on your physical android device. Keep your device plugged in and open Logcat, in the filter at the top type in "Batch size" if you want to find the error easily.

Step 7. Once the app opens, swipe up on the arrow at the bottom to open the selection menu, from here set ML Model to Selfie Multiclass and then set the Delegate to GPU

Step 8. View the error thrown on the screen and in Logcat

2025-05-30 09:21:01.266  4379-4438  native                  com...pe.examples.imagesegmentation  E  E0000 00:00:1748611261.266161    4438 calculator_graph.cc:887] INVALID_ARGUMENT: CalculatorGraph::Run() failed: 
                                                                                                    Calculator::Open() for node "mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: Batch size mismatch, expected 1 but got 8
2025-05-30 09:21:01.267  4379-4438  ImageSegmenterHelper    com...pe.examples.imagesegmentation  E  Image segmenter failed to load model with error: invalid argument: CalculatorGraph::Run() failed: 
                                                                                                    Calculator::Open() for node "mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_image_segmenter_imagesegmentergraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: Batch size mismatch, expected 1 but got 8

[kuaashish]

Hi @peymannaderi10,

Thanks for the report. We have confirmed and reproduced the crash: selecting 'GPU' as the delegate and 'Selfie Multiclass' as the ML Model leads to a crash. All other models tested work as expected in Pixel 6a. We will highlight this to right point of contact. Hope it will be fixed soon.