Community Contributed Model

Overview

The Community Contributed Model section collects benchmarking information for models contributed by developers and partners worldwide. These entries are intended for evaluation and reference only and may not provide a complete end-to-end workflow.

In particular:

• Source training code and original datasets are typically not provided.

• Conversion steps from the original framework to deployment formats (for example, TFLite or DLA) may be undocumented.

• Only inference performance and basic configuration details are recorded.

Use these models primarily to understand achievable performance on MediaTek Genio platforms and as inspiration for building your own optimized pipelines.

Benchmark Results

The following table summarizes performance results for community‑contributed models:

Model Name	Source model type	Data Type	Input Size	Genio 510	Genio 700	Genio 1200	Genio 520	Gneio 720
MobileNetV1	.tflite	Quant8	224x224	1.28	1.04	1.05	0.93	0.92
YOLOv11s-quant8	.tflite	Quant8	640x640	27.04	19.04	Not Supported	8.56	8.52
YOLOv11s-FP32	.tflite	FP32	640x640	72.03	50.04	Not Supported	24.94	24.93

Note

N/A entries in the table indicate that the data is unavailable because the corresponding TFLite model contains the BATCH_MATMUL operator, which MDLA 2.0 does not support.