ResNet-2Plus1D: Optimized for Mobile Deployment

Sports and human action recognition in videos

ResNet (2+1)D Convolutions is a network which explicitly factorizes 3D convolution into two separate and successive operations, a 2D spatial convolution and a 1D temporal convolution. It used for video understanding applications.

This model is an implementation of ResNet-2Plus1D found here.

This repository provides scripts to run ResNet-2Plus1D on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Model_use_case.video_classification
Model Stats:
- Model checkpoint: Kinetics-400
- Input resolution: 112x112
- Number of parameters: 31.5M
- Model size (float): 120 MB
- Model size (w8a8): 30.8 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
ResNet-2Plus1D	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	736.652 ms	0 - 195 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	81.891 ms	0 - 181 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	459.478 ms	0 - 272 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	26.954 ms	2 - 247 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	398.418 ms	0 - 3 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	12.863 ms	2 - 4 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	11.931 ms	0 - 64 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	392.239 ms	0 - 195 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	21.602 ms	0 - 192 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	736.652 ms	0 - 195 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	81.891 ms	0 - 181 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	409.844 ms	0 - 4 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	12.839 ms	2 - 5 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	475.165 ms	0 - 188 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	22.955 ms	0 - 168 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	396.099 ms	0 - 3 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	12.838 ms	2 - 4 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	392.239 ms	0 - 195 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	21.602 ms	0 - 192 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	292.093 ms	0 - 281 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	9.306 ms	2 - 269 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	8.731 ms	2 - 233 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	266.815 ms	0 - 193 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	7.356 ms	2 - 184 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	7.199 ms	1 - 144 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	259.756 ms	0 - 203 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	5.418 ms	2 - 191 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	5.51 ms	2 - 147 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	13.285 ms	2 - 2 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	12.322 ms	60 - 60 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Dragonwing Q-6690 MTP	Qualcomm® Qcm6690	TFLITE	1621.204 ms	321 - 478 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Dragonwing Q-6690 MTP	Qualcomm® Qcm6690	QNN_DLC	68.331 ms	1 - 167 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Dragonwing Q-6690 MTP	Qualcomm® Qcm6690	ONNX	299.164 ms	99 - 112 MB	CPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	TFLITE	1797.052 ms	267 - 428 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	QNN_DLC	17.317 ms	1 - 3 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	ONNX	317.409 ms	98 - 129 MB	CPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	1292.279 ms	0 - 394 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	13.087 ms	1 - 164 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	919.714 ms	0 - 371 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	7.175 ms	1 - 202 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	699.325 ms	0 - 3 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	4.426 ms	1 - 3 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	4.359 ms	0 - 35 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	672.511 ms	0 - 339 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	4.653 ms	1 - 163 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	RB5 (Proxy)	Qualcomm® QCS8250 (Proxy)	TFLITE	1302.786 ms	407 - 439 MB	CPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	RB5 (Proxy)	Qualcomm® QCS8250 (Proxy)	ONNX	281.369 ms	71 - 124 MB	CPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	SA7255P ADP	Qualcomm® SA7255P	TFLITE	1292.279 ms	0 - 394 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	13.087 ms	1 - 164 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	700.376 ms	0 - 3 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	4.435 ms	1 - 2 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	SA8295P ADP	Qualcomm® SA8295P	TFLITE	763.639 ms	0 - 340 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	7.567 ms	1 - 166 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	700.45 ms	0 - 3 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	4.433 ms	1 - 3 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	SA8775P ADP	Qualcomm® SA8775P	TFLITE	672.511 ms	0 - 339 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	4.653 ms	1 - 163 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	512.136 ms	0 - 438 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	3.222 ms	1 - 204 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	3.205 ms	0 - 184 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	396.944 ms	0 - 419 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	2.411 ms	1 - 161 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	2.615 ms	0 - 130 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	TFLITE	1244.845 ms	334 - 510 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	QNN_DLC	7.607 ms	1 - 158 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	ONNX	286.964 ms	95 - 110 MB	CPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	582.234 ms	0 - 372 MB	NPU	ResNet-2Plus1D.tflite
ResNet-2Plus1D	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	1.782 ms	1 - 165 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	1.906 ms	0 - 131 MB	NPU	ResNet-2Plus1D.onnx.zip
ResNet-2Plus1D	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	4.749 ms	1 - 1 MB	NPU	ResNet-2Plus1D.dlc
ResNet-2Plus1D	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	4.487 ms	31 - 31 MB	NPU	ResNet-2Plus1D.onnx.zip

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[resnet-2plus1d]"

Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub Workbench with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.resnet_2plus1d.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.resnet_2plus1d.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.resnet_2plus1d.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.resnet_2plus1d import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. Sign up for access.

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on ResNet-2Plus1D's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of ResNet-2Plus1D can be found here.

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.

Downloads last month: 60

Inference Providers NEW

Video Classification

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for qualcomm/ResNet-2Plus1D

A Closer Look at Spatiotemporal Convolutions for Action Recognition

Paper • 1711.11248 • Published Nov 30, 2017