Pre-built yolov8 demo for Alveo u55c

finn_examples/models.py

@@ -34,6 +34,7 @@
 from qonnx.core.datatype import DataType
 
 from finn_examples.driver import FINNExampleOverlay
+from finn_examples.yolov8.yolov8 import DetectorDriver
 
 _mnist_fc_io_shape_dict = {
     "idt": [DataType["UINT8"]],
@@ -99,6 +100,27 @@
     "num_outputs": 1,
 }
 
+_yolov8n_4w4a_coco_io_shape_dict = {
+    # FINN DataType for input and output tensors
+    "idt": [DataType["UINT8"]],
+    "odt": [DataType["INT21"], DataType["INT21"], DataType["INT21"]],
+    # shapes for input and output tensors (NHWC layout)
+    "ishape_normal": [(1, 192, 320, 3)],
+    "oshape_normal": [(1, 24, 40, 144), (1, 12, 20, 144), (1, 6, 10, 144)],
+    # folded / packed shapes below depend on idt/odt and input/output
+    # PE/SIMD parallelization settings -- these are calculated by the
+    # FINN compiler.
+    "ishape_folded": [(1, 192, 320, 3, 1)],
+    "oshape_folded": [(1, 24, 40, 144, 1), (1, 12, 20, 144, 1), (1, 6, 10, 144, 1)],
+    "ishape_packed": [(1, 192, 320, 3, 1)],
+    "oshape_packed": [(1, 24, 40, 144, 3), (1, 12, 20, 144, 3), (1, 6, 10, 144, 3)],
+    "input_dma_name": ["idma0"],
+    "output_dma_name": ["odma0", "odma1", "odma2"],
+    "number_of_external_weights": 0,
+    "num_inputs": 1,
+    "num_outputs": 3,
+}
+
 _imagenet_top5inds_io_shape_dict = {
     "idt": [DataType["UINT8"]],
     "odt": [DataType["UINT16"]],
@@ -333,6 +355,17 @@ def bincop_cnv(target_platform=None, bitfile_path=None):
     return FINNExampleOverlay(filename, driver_mode, _bincop_cnv_io_shape_dict)
 
 
+def yolov8n_4w4a_coco(target_platform=None, bitfile_path=None, batch_size=1):
+    target_platform = resolve_target_platform(target_platform)
+    driver_mode = get_driver_mode()
+    model_name = "yolov8n_4w4a_coco"
+    filename = find_bitfile(model_name, target_platform, bitfile_path)
+    quant_tail_params_dir = pk.resource_filename("finn_examples", "yolov8")
+    return DetectorDriver(
+        filename, driver_mode, _yolov8n_4w4a_coco_io_shape_dict, quant_tail_params_dir, batch_size
+    )
+
+
 def mobilenetv1_w4a4_imagenet(target_platform=None, bitfile_path=None, rt_weights_path=None):
     target_platform = resolve_target_platform(target_platform)
     driver_mode = get_driver_mode()

finn_examples/notebooks/8_yolov8_object_detection.ipynb

@@ -0,0 +1,86 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import cv2\n",
+    "from finn_examples import models"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Initialize the Accelerator"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# The yolov8 is pre-build only for alveo u55c at the moment\n",
+    "accel = models.yolov8n_4w4a_coco()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Run Inference"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "img = cv2.imread('images/test.jpg')\n",
+    "detections = accel.single_inference(img, conf_thres=0.3) "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Visualize output"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "visualized_img = accel.visualize(img, detections)\n",
+    "cv2.imwrite('demo.jpg', visualized_img)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 1
+}