Add model

README.md

@@ -15,8 +15,6 @@ tags:
 
 # CenterNet-R18 (ONNX) – Renesas X5H
 
-> ⏳ **Model file not yet uploaded.** Benchmark results on this page were published ahead of the model weights — see **Provided Artifacts** below. Download/deployment steps will not work until the file is added to this repository.
-
 ## Introduction
 
 This repository hosts **CenterNet** with a **ResNet18** backbone, targeting the **Renesas
@@ -45,7 +43,7 @@ centernet_r18_..._optimized.onnx (FP32)
 
 | Artifact | Status | Notes |
 |----------|--------|-------|
-| **FP32 (ONNX)** | ⏳ Not yet uploaded | Benchmark numbers below exist; the model file has not been published to this repo yet |
+| **FP32 (ONNX)** | ✅ Published | `fp32/centernet_r18_8xb16_crop512_140e_coco.onnx` — auto-cast to INT8 by the MWMX toolchain at compile time (see Deployment Flow above); no separate INT8 file is shipped |
 
 ## Performance
 
@@ -56,6 +54,7 @@ Measured on **Renesas R-Car X5H** via the MWMX runtime (APM80 ship-performance C
 | Runtime | Precision | Device | Latency (ms) | Type |
 |---------|-----------|--------|---------------|------|
 | MWMX Runtime | INT8 (auto) | X5H · 1× NPU · 1 Core · 850 MHz | 11.978 | Measured |
+| MWMX Runtime | INT8 (auto) | X5H · 1× NPU · 1 Core · 850 MHz | 11.989 | Measured (2026-09-16) |
 | MWMX Runtime | INT8 (auto) | X5H · 1× NPU · 12 Cores · 850 MHz | 3.467 | Measured |
 
 ### Accuracy
@@ -81,11 +80,13 @@ To run inference on Renesas R-Car X5H, you need:
 
 1. **Renesas R-Car X5H board** with NPX6 NPU
 2. **Renesas MWMX Runtime**
-3. **Hugging Face CLI** to download the model (once the model file is published)
+3. **Hugging Face CLI** to download the model
 
 ## Download
 
-TBD — model file not yet published to this repository.
+```bash
+hf download Renesas/CenterNet-R18-ONNX --repo-type=model --include "fp32/*"
+```
 
 ---
 

compile_config/l2seg_csm_16mb.json

@@ -0,0 +1,69 @@
+{
+  "segments": [
+    {
+      "name": "Segment_0",
+      "inputs": [
+        "input"
+      ],
+      "outputs": [
+        "/neck/deconv_layers/deconv_layers.5/activate/Relu_output_0"
+      ]
+    },
+    {
+      "name": "Segment_4",
+      "inputs": [
+        "/neck/deconv_layers/deconv_layers.5/activate/Relu_output_0"
+      ],
+      "outputs": [
+        "/bbox_head/Sigmoid_output_0"
+      ]
+    },
+    {
+      "name": "Segment_5",
+      "inputs": [
+        "/neck/deconv_layers/deconv_layers.5/activate/Relu_output_0"
+      ],
+      "outputs": [
+        "/bbox_head/wh_head/wh_head.2/Conv_output_0"
+      ]
+    },
+    {
+      "name": "Segment_6",
+      "inputs": [
+        "/neck/deconv_layers/deconv_layers.5/activate/Relu_output_0"
+      ],
+      "outputs": [
+        "/bbox_head/offset_head/offset_head.2/Conv_output_0"
+      ]
+    }
+  ],
+  "layer_groups": [
+    {
+      "name": "LG0",
+      "inputs": [
+        "/backbone/maxpool/MaxPool_output_0"
+      ],
+      "outputs": [
+        "/backbone/layer1/layer1.0/relu_1/Relu_output_0"
+      ]
+    },
+    {
+      "name": "LG1",
+      "inputs": [
+        "/backbone/layer1/layer1.0/relu_1/Relu_output_0"
+      ],
+      "outputs": [
+        "/backbone/layer1/layer1.1/relu_1/Relu_output_0"
+      ]
+    },
+    {
+      "name": "LG2",
+      "inputs": [
+        "/backbone/layer1/layer1.1/relu_1/Relu_output_0"
+      ],
+      "outputs": [
+        "/backbone/layer2/layer2.0/conv2/Conv_output_0"
+      ]
+    }
+  ]
+}

fp32/.metadata.yaml

@@ -0,0 +1,5 @@
+variant:
+  id: onnx_fp32
+  format: onnx
+  precision: fp32
+  method: onnx_export        # exported straight from the OpenMMLab checkpoint, no quantization

fp32/centernet_r18_8xb16_crop512_140e_coco.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:279c1b2d971c9d2488b137a6115805115c4c875f505b160d8ff968031152f888
+size 56884157

int8/benchmarks/x5h_mwmx_npu_apm80_12core.yaml

@@ -40,3 +40,51 @@ memory:
 
 power:
   avg_w: null
+
+# Exact commands verified against the NNAC "Getting Started" chapter for this
+# model (it is that chapter's own worked example), plus the network_configuration
+# block verified against the internal model_zoo compile config for this model's
+# 12-core slice (see compile_config/l2seg_csm_16mb.json). Rendered by the
+# AI-Dashboard in place of the generic placeholder flow — see
+# downloadRunHTML() / parse_reproduce(). NOTE: the 1-core vs. 12-core split
+# (see `configuration` above) is set at compile time via
+# ${WORKDIR}/nnac_config/config.yaml, not a host_app flag — the on-board run
+# command below is identical to the 1-core run.
+reproduce:
+  steps:
+    - title: Download the ONNX model and compile config
+      command: hf download Renesas/CenterNet-R18-ONNX --repo-type=model --include "fp32/*" --include "compile_config/*"
+    - title: Configure the network for the 12-core build
+      command: |
+        cat >> ${WORKDIR}/nnac_config/config.yaml << 'EOF'
+        network_configuration:
+          centernet_r18_8xb16_crop512_140e_coco:
+            last_nodes:
+              - /Reshape_output_0
+              - /Flatten_output_0
+              - /Reshape_1_output_0
+            global_config:
+              weight_symmetric: true
+            optimization_options:
+              - -segmentation_json ./compile_config/l2seg_csm_16mb.json
+              - -enable_shallow_conv_opt
+              - -enable_dma_fusion
+              - -stu_weight
+              - -distributed_coef_stu
+        EOF
+        # If config.yaml already has a network_configuration: key (e.g. from a
+        # previous model), merge this block into it instead of appending a duplicate key.
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./fp32/centernet_r18_8xb16_crop512_140e_coco.onnx
+    - title: Set up the R-Car X5H board
+      command: Configure the board per the AI Compiler (NNAC) "Getting Started" guide, section 3.4 (host TFTP/NFS setup, bootloader flashing, U-Boot, Linux boot, login) -- exact steps depend on your board/network setup.
+    - title: Copy the compiled artifact to the board
+      command: Copy ${WORKDIR}/binary/nnx to the board -- method may vary (NFS mount, scp, USB, etc.).
+    - title: Run on R-Car X5H (single NPU cluster, 12 AI cores)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/centernet_r18_8xb16_crop512_140e_coco
+  notes: >-
+    hash[n] = 0x...(OK) means the run's output matches the reference hash in
+    hash.txt; latency is the NPX execution time reported in cycles and ms.

int8/benchmarks/x5h_mwmx_npu_apm80_1core.yaml

@@ -40,3 +40,25 @@ memory:
 
 power:
   avg_w: null
+
+# Exact commands verified against the NNAC "Getting Started" chapter for this
+# model (it is that chapter's own worked example). Rendered by the AI-Dashboard
+# in place of the generic placeholder flow — see downloadRunHTML() / parse_reproduce().
+reproduce:
+  steps:
+    - title: Download the ONNX model
+      command: hf download Renesas/CenterNet-R18-ONNX --repo-type=model --include "fp32/*"
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./fp32/centernet_r18_8xb16_crop512_140e_coco.onnx
+    - title: Set up the R-Car X5H board
+      command: Configure the board per the AI Compiler (NNAC) "Getting Started" guide, section 3.4 (host TFTP/NFS setup, bootloader flashing, U-Boot, Linux boot, login) -- exact steps depend on your board/network setup.
+    - title: Copy the compiled artifact to the board
+      command: Copy ${WORKDIR}/binary/nnx to the board -- method may vary (NFS mount, scp, USB, etc.).
+    - title: Run on R-Car X5H (single NPU cluster)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/centernet_r18_8xb16_crop512_140e_coco
+  notes: >-
+    hash[n] = 0x...(OK) means the run's output matches the reference hash in
+    hash.txt; latency is the NPX execution time reported in cycles and ms.

int8/benchmarks/x5h_mwmx_npu_apm80_1core_2026-09-16.yaml

@@ -0,0 +1,65 @@
+# Re-run captured 2026-09-16 (gf_model_result.csv export) — see also the original file in this directory
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: mwmx
+  toolchain_version: "MWMX SDK v4.35.0"
+  format: onnx
+  execution_provider: npu
+  execution_precision: int8
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 1
+  npu_freq_mhz: 850
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 512, 512]   # inferred from "crop512" in the source checkpoint name
+
+performance:
+  fps: null
+  latency: 11.989098        # APM80 pipeline run (metawaremx_runtime CI)
+
+metrics:
+  accuracy: null
+  top5_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+# Exact commands verified against the NNAC "Getting Started" chapter for this
+# model (it is that chapter's own worked example). Rendered by the AI-Dashboard
+# in place of the generic placeholder flow — see downloadRunHTML() / parse_reproduce().
+reproduce:
+  steps:
+    - title: Download the ONNX model
+      command: hf download Renesas/CenterNet-R18-ONNX --repo-type=model --include "fp32/*"
+    - title: Compile with the NNAC toolchain (INT8 auto-cast from the FP32 graph)
+      command: |
+        python3 nnac_frontend/legalize.py -d binary/nnx ./fp32/centernet_r18_8xb16_crop512_140e_coco.onnx
+    - title: Set up the R-Car X5H board
+      command: Configure the board per the AI Compiler (NNAC) "Getting Started" guide, section 3.4 (host TFTP/NFS setup, bootloader flashing, U-Boot, Linux boot, login) -- exact steps depend on your board/network setup.
+    - title: Copy the compiled artifact to the board
+      command: Copy ${WORKDIR}/binary/nnx to the board -- method may vary (NFS mount, scp, USB, etc.).
+    - title: Run on R-Car X5H (single NPU cluster)
+      command: |
+        cd binary
+        ./host_app ./arc_prog_npus ./nnx/centernet_r18_8xb16_crop512_140e_coco
+  notes: >-
+    hash[n] = 0x...(OK) means the run's output matches the reference hash in
+    hash.txt; latency is the NPX execution time reported in cycles and ms.