LiteRT is Google's on-device runtime, the new name for TensorFlow Lite (Android: com.google.ai.edge.litert:litert), and litert-torch, the renamed ai-edge-torch, is its PyTorch converter: a PyTorch model converted unmodified with litert_torch.convert matched the original to 4e-7 on a Galaxy S26 (measured, LiteRT 2.2.0, Android 16, 2026-09-05).

PIDNet-S — LiteRT (real-time semantic segmentation, GPU)

On-device real-time semantic segmentation running fully on the LiteRT CompiledModel GPU delegate (no CPU fallback). PIDNet-S (CVPR 2023) segments a road scene into the 19 Cityscapes classes at ~17 FPS on a Pixel 8a.

Architecture: PIDNet-S — a three-branch CNN (P: detail, I: context, D: boundary).
Weights: XuJiacong/PIDNet · MIT · 78.8% mIoU (Cityscapes val).
Size: 30 MB · ~7.6 M params · pure CNN.

I/O

Input: [1, 3, 1024, 1024] NCHW, RGB, ImageNet-normalized (mean [0.485,0.456,0.406], std [0.229,0.224,0.225]).
Output: [1, 19, 128, 128] class logits at 1/8 resolution — argmax over the 19 classes per pixel, then upscale (nearest) to display.

Classes (index order):

road, sidewalk, building, wall, fence, pole, traffic light, traffic sign, vegetation, terrain, sky, person, rider, car, truck, bus, train, motorcycle, bicycle

GPU conversion

PIDNet is a pure CNN — no attention, no dynamic shapes at a fixed input size, and align_corners=False on every bilinear resize. It converts to a fully GPU-compatible graph with zero patches: CONV_2D ×75, RESIZE_BILINEAR ×11 (align_corners=False), AVERAGE_POOL_2D, ADD/MUL/SUB/SUM, LOGISTIC — 0 tensors of rank > 4, 0 GPU-incompatible ops. The converted graph matches the original PyTorch model bit-for-bit on CPU (corr 0.99999999999, 100% argmax); on the Mali GPU (fp16) it agrees with the fp32 reference at 97% of pixels with correct classes.

Minimal usage

Kotlin (Android, LiteRT CompiledModel GPU)

kotlin

val options = CompiledModel.Options(Accelerator.GPU)
val model = CompiledModel.create(context.assets, "pidnet_s.tflite", options, null)
val inBufs = model.createInputBuffers()
val outBufs = model.createOutputBuffers()

inBufs[0].writeFloat(inputNCHW)              // [1,3,1024,1024], RGB, ImageNet-norm
model.run(inBufs, outBufs)
val logits = outBufs[0].readFloat()          // [19,128,128] (NCHW, batch dropped)

// argmax over 19 classes per pixel:
val hw = 128 * 128
val label = IntArray(hw) { i ->
    var best = 0; var bv = logits[i]
    for (c in 1 until 19) { val v = logits[c * hw + i]; if (v > bv) { bv = v; best = c } }
    best
}

Python (LiteRT / ai-edge-litert)

python

from ai_edge_litert.interpreter import Interpreter
import numpy as np

it = Interpreter(model_path="pidnet_s.tflite"); it.allocate_tensors()
inp, out = it.get_input_details(), it.get_output_details()
it.set_tensor(inp[0]["index"], x)            # [1,3,1024,1024] float32, ImageNet-norm
it.invoke()
logits = it.get_tensor(out[0]["index"])[0]   # [19,128,128]
label = logits.argmax(0)                      # [128,128] class ids

Conversion

Re-authored/converted with litert-torch (build_pidnet.py): the trained PIDNet-S weights are loaded from an ONNX mirror whose initializer names match the original repo's PyTorch keys, then converted directly — zero GPU patches.

Performance

Measured on a Pixel 8a (Tensor G3, Android 16) with the standard TFLite benchmark_model tool — 10 warm-up runs then 50 timed runs, reported as the tool's mean.

Runtime	Backend	Graph on GPU	Latency
TFLite `benchmark_model` (`TfLiteGpuDelegateV2`)	GPU (OpenCL)	190 / 190	61.5 ms
TFLite `benchmark_model`	CPU (XNNPACK, 4 threads)	—	719.2 ms

Any on-device figure recorded when this model shipped came from a different runtime. It was taken through LiteRT's own CompiledModel accelerator (logcat reports it as LITERT_CL), which is the path the Kotlin sample app and the LiteRT API use, and it appears elsewhere on this card. The rows above are the classic TFLite OpenCL delegate, measured with a tool anyone can download and re-run. The two are not comparable, so read the rows above as a reproducible floor rather than as this model's speed on LiteRT.

Snapdragon NPU (Hexagon)

The NPU is 2.95x faster than the GPU (5.48 ms against 16.20 ms) and loads 12.62x faster (119 ms against 1502 ms).

backend	inference (median / min)	load
NPU (Hexagon v81)	5.48 ms / 5.44 ms	119 ms
GPU (Adreno)	16.20 ms / 15.73 ms	1502 ms

Measured on a Samsung Galaxy S26 (Snapdragon 8 Elite Gen 5 / SM8850, Hexagon v81, Android 16), LiteRT CompiledModel 2.2.0, one accelerator per process, 5 warm-up runs then N=50 timed runs, median reported. Every run held thermal status NONE throughout. Headroom 0.67, where 1.0 is the throttling threshold.

The NPU rows here ran artifacts compiled ahead of time for SM8850 with QAIRT 2.47.0; the GPU rows ran the published files as they are. LiteRT can also compile for the NPU on the device at first load, which is what lets you ship the published file unchanged — that path and the ten runtime libraries it needs are in the NPU recipe, and we did not measure it here. GPU wiring is in the GPU recipe.

Raspberry Pi 5 (CPU)

Measured on a Raspberry Pi 5 Model B Rev 1.1 (8 GB, Raspberry Pi OS 64-bit) with the LiteRT benchmark_model tool from litert-cli-nightly 0.2.0.dev20260805: CPU inference (XNNPACK, 4 threads), 3 invocations per file of 10 warm-up plus 50 timed runs (the tool caps a phase at 150 s, so very slow graphs run fewer — the Runs column is the actual timed total). The latency is the median across invocations; the spread is the min–max over all timed runs. No thermal throttling occurred during these runs (vcgencmd get_throttled stayed 0x0).

File	Inference (median)	Spread (min–max)	Runs	Peak memory
`pidnet_s.tflite`	482.3 ms	480.3–485.6 ms	150	201 MB

License

MIT (PIDNet / XuJiacong/PIDNet). Cityscapes label taxonomy from the Cityscapes dataset.

litert-community/PIDNet-S-Cityscapes-LiteRT