litert-community/RTMPose-s-LiteRT

Measured on device (edge-compat): Galaxy S26 · LiteRT 2.2.0 · GPU (ML Drift) · 3.36 ms p50 (2026-08-25); Galaxy S26 · LiteRT 2.2.0 · NPU (QNN/HTP) · 0.77 ms p50 (2026-08-25); Raspberry Pi 5 · LiteRT 2.2.0.dev20260804 · CPU/XNNPACK, 4 threads · 16.7 ms p50 (2026-08-31); browser · Chromium 151 on M4 Max · LiteRT.js 2.5.3 · WebGPU · 2.96 ms p50 · output matches CPU (2026-08-11). Record: https://github.com/john-rocky/edge-compat/blob/main/cards/rtmpose-s/CARD.md

RTMPose-s — LiteRT (on-device real-time 2D human pose, fully-GPU)

RTMPose (mmpose, CSPNeXt backbone + RTMCC/SimCC head) top-down 2D human pose, converted to LiteRT and running fully on the CompiledModel GPU (ML Drift) on Android. Estimates 17 COCO keypoints for a single centered person — the SOTA real-time pose model, device-verified end-to-end.

On-device (Pixel 8a, Tensor G3 — verified)

nodes on GPU	256 / 256 LITERT_CL (full residency)
inference	~4 ms (256×192)
size	11.1 MB (fp16)
accuracy	device-vs-PyTorch SimCC corr 0.999, keypoints within 0.3 px (max 1 px)

image[1,3,256,192] (ImageNet 0-255 norm) →[GPU: CSPNeXt + RTMCC]→ simcc_x[1,17,384], simcc_y[1,17,512]

The SimCC head emits two 1D distributions per keypoint; argmax over the bins (÷ split=2) gives the pixel x/y.

Minimal usage

Android (Kotlin, CompiledModel GPU)

val model = CompiledModel.create(context.assets, "rtmpose_s_fp16.tflite",
    CompiledModel.Options(Accelerator.GPU), null)
val inputs = model.createInputBuffers()
val outputs = model.createOutputBuffers()
inputs[0].writeFloat(chw)              // [1,3,256,192] mmpose mean/std (0-255 RGB), NCHW
model.run(inputs, outputs)
val simccX = outputs[0].readFloat()    // [1,17,384]
val simccY = outputs[1].readFloat()    // [1,17,512]; keypoint = argmax / 2

Python (desktop verification)

MEAN = np.array([123.675, 116.28, 103.53], np.float32)
STD  = np.array([58.395, 57.12, 57.375], np.float32)
import numpy as np
from PIL import Image
from ai_edge_litert.interpreter import Interpreter

img = Image.open("person.jpg").convert("RGB").resize((192, 256))  # centered subject crop
x = ((np.asarray(img, np.float32) - MEAN) / STD).transpose(2, 0, 1)[None]

it = Interpreter(model_path="rtmpose_s_fp16.tflite"); it.allocate_tensors()
it.set_tensor(it.get_input_details()[0]["index"], x); it.invoke()
od = it.get_output_details()
sx, sy = (it.get_tensor(o["index"])[0] for o in od)              # [17,384], [17,512]
if sx.shape[-1] != 384: sx, sy = sy, sx                          # identify by bin count
kx, ky = sx.argmax(-1) / 2.0, sy.argmax(-1) / 2.0                 # 17 keypoints, px in 192x256
for i, (a, b) in enumerate(zip(kx, ky)):
    print(f"kp{i}: ({a:.1f}, {b:.1f})")

How it converts (litert-torch) — two numerically-exact re-authorings

Both are on-device-only Mali issues: they pass the desktop op-check and report full LITERT_CL residency, yet the device output was wrong until fixed (residency ≠ correctness):

1. ScaleNorm (RMS norm) fp16 overflow → all-zero head. The RTMCC ScaleNorm input reaches ≈ |274|, so its channel Σ x² ≈ 3.6M overflows fp16 (max 65504) on the Mali delegate (which reduces in fp16 even for an fp32 graph) → norm = ∞ → x/∞ = 0 → the whole head collapses to zero. Fix: scale x down by S=64 before squaring, then rescale (math-identical) — a SafeRMSNorm.
2. GAU attention act@act BMM → broadcast-reduce. The Gated Attention Unit's q@kᵀ and kernel@v are activation×activation batch-matmuls that the Mali delegate mis-computes; at K=17 tokens the exact replacement is (q[:,:,None,:]·k[:,None,:,:]).sum(-1).

Result: banned ops NONE, all tensors ≤4D, tflite-vs-torch corr 1.0, device-vs-torch corr 0.999.

Preprocessing

Center-crop to 3:4, resize to 192×256, ImageNet 0-255 normalize (mean [123.675, 116.28, 103.53], std [58.395, 57.12, 57.375]), NCHW planar. Top-down — expects one roughly-centered person.

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
LiteRT CompiledModel (LITERT_CL)	GPU	256 / 256	~4 ms
TFLite benchmark_model (TfLiteGpuDelegateV2)	GPU (OpenCL)	256 / 256	15.8 ms
TFLite benchmark_model	CPU (XNNPACK, 4 threads)	—	XNNPACK declined the graph

The two GPU rows are different runtimes, not a contradiction. The LITERT_CL figure is the one recorded when this model shipped, taken through LiteRT's own CompiledModel accelerator — the path the Kotlin sample app and the LiteRT API use. The TfLiteGpuDelegateV2 figure is the classic TFLite OpenCL delegate, measured with a tool anyone can download and re-run. They agree on how much of the graph the GPU takes; they disagree on speed, and the classic delegate is the slower of the two here. Read the TfLiteGpuDelegateV2 row as a reproducible floor, not as this model's speed on LiteRT.

XNNPACK declines these fp16 graphs — it reports failed to delegate DEPTHWISE_CONV_2D and then fails to allocate tensors — so there is no usable CPU number. Disabling XNNPACK falls back to reference kernels, which measured about 20× slower than the GPU on models of this size and would not represent CPU inference anyone would ship.

Snapdragon NPU (Hexagon)

The NPU is 4.36x faster than the GPU (0.771 ms against 3.36 ms) and loads 10.12x faster (102 ms against 1033 ms).

backend	compiled	inference (median / min)	load
NPU (Hexagon v81)	on-device JIT	0.771 ms / 0.749 ms	102 ms
GPU (Adreno)	—	3.36 ms / 2.85 ms	1033 ms

Measured on a Samsung Galaxy S26 (Snapdragon 8 Elite Gen 5 / SM8850, Hexagon v81, Android 16) with 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 0.79, where 1.0 is the throttling threshold.

The NPU rows ran the published file unchanged. LiteRT compiled it for the Hexagon on the device at first load. That first compile took 533 ms here. The load column above is the cached load every later run pays. Recipe and the runtime libraries it needs: NPU guide.

GPU wiring: GPU guide.

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
rtmpose_s_fp16.tflite	16.7 ms	16.4–19.1 ms	181	138 MB

License

Apache-2.0. Upstream: open-mmlab/mmpose RTMPose-s.