Compile YOLO-V2 and YOLO-V3 in DarkNet Models

Author: Siju Samuel

This article is an introductory tutorial to deploy darknet models with TVM. All the required models and libraries will be downloaded from the internet by the script. This script runs the YOLO-V2 and YOLO-V3 Model with the bounding boxes Darknet parsing have dependancy with CFFI and CV2 library Please install CFFI and CV2 before executing this script

pip install cffi
pip install opencv-python
# numpy and matplotlib
import numpy as np
import matplotlib.pyplot as plt
import sys

# tvm, relay
import tvm
from tvm import relay
from ctypes import *
from import download_testdata
from tvm.relay.testing.darknet import __darknetffi__
import tvm.relay.testing.yolo_detection
import tvm.relay.testing.darknet

# Model name
MODEL_NAME = 'yolov3'

Download required files

Download cfg and weights file if first time.

CFG_URL = REPO_URL + 'cfg/' + CFG_NAME + '?raw=true'

cfg_path = download_testdata(CFG_URL, CFG_NAME, module="darknet")
weights_path = download_testdata(WEIGHTS_URL, WEIGHTS_NAME, module="darknet")

# Download and Load darknet library
if sys.platform in ['linux', 'linux2']:
    DARKNET_LIB = ''
    DARKNET_URL = REPO_URL + 'lib/' + DARKNET_LIB + '?raw=true'
elif sys.platform == 'darwin':
    DARKNET_LIB = ''
    DARKNET_URL = REPO_URL + 'lib_osx/' + DARKNET_LIB + '?raw=true'
    err = "Darknet lib is not supported on {} platform".format(sys.platform)
    raise NotImplementedError(err)

lib_path = download_testdata(DARKNET_URL, DARKNET_LIB, module="darknet")

DARKNET_LIB = __darknetffi__.dlopen(lib_path)
net = DARKNET_LIB.load_network(cfg_path.encode('utf-8'), weights_path.encode('utf-8'), 0)
dtype = 'float32'
batch_size = 1

data = np.empty([batch_size, net.c, net.h, net.w], dtype)
shape_dict = {'data': data.shape}
print("Converting darknet to relay functions...")
sym, params = relay.frontend.from_darknet(net, dtype=dtype, shape=data.shape)


File /workspace/.tvm_test_data/darknet/yolov3.cfg exists, skip.
File /workspace/.tvm_test_data/darknet/yolov3.weights exists, skip.
File /workspace/.tvm_test_data/darknet/ exists, skip.
Converting darknet to relay functions...

Import the graph to Relay

compile the model

target = 'llvm'
target_host = 'llvm'
ctx = tvm.cpu(0)
data = np.empty([batch_size, net.c, net.h, net.w], dtype)
shape = {'data': data.shape}
print("Compiling the model...")
with relay.build_config(opt_level=3):
    graph, lib, params =, target=target, target_host=target_host, params=params)

[neth, netw] = shape['data'][2:] # Current image shape is 608x608


Compiling the model...

Load a test image

test_image = 'dog.jpg'
print("Loading the test image...")
img_url = REPO_URL + 'data/' + test_image + '?raw=true'
img_path = download_testdata(img_url, test_image, "data")

data = tvm.relay.testing.darknet.load_image(img_path, netw, neth)


Loading the test image...
File /workspace/.tvm_test_data/data/dog.jpg exists, skip.

Execute on TVM Runtime

The process is no different from other examples.

from tvm.contrib import graph_runtime

m = graph_runtime.create(graph, lib, ctx)

# set inputs
m.set_input('data', tvm.nd.array(data.astype(dtype)))
# execute
print("Running the test image...")
# get outputs
tvm_out = []
if MODEL_NAME == 'yolov2':
    layer_out = {}
    layer_out['type'] = 'Region'
    # Get the region layer attributes (n, out_c, out_h, out_w, classes, coords, background)
    layer_attr = m.get_output(2).asnumpy()
    layer_out['biases'] = m.get_output(1).asnumpy()
    out_shape = (layer_attr[0], layer_attr[1]//layer_attr[0],
                 layer_attr[2], layer_attr[3])
    layer_out['output'] = m.get_output(0).asnumpy().reshape(out_shape)
    layer_out['classes'] = layer_attr[4]
    layer_out['coords'] = layer_attr[5]
    layer_out['background'] = layer_attr[6]

elif MODEL_NAME == 'yolov3':
    for i in range(3):
        layer_out = {}
        layer_out['type'] = 'Yolo'
        # Get the yolo layer attributes (n, out_c, out_h, out_w, classes, total)
        layer_attr = m.get_output(i*4+3).asnumpy()
        layer_out['biases'] = m.get_output(i*4+2).asnumpy()
        layer_out['mask'] = m.get_output(i*4+1).asnumpy()
        out_shape = (layer_attr[0], layer_attr[1]//layer_attr[0],
                     layer_attr[2], layer_attr[3])
        layer_out['output'] = m.get_output(i*4).asnumpy().reshape(out_shape)
        layer_out['classes'] = layer_attr[4]

# do the detection and bring up the bounding boxes
thresh = 0.5
nms_thresh = 0.45
img = tvm.relay.testing.darknet.load_image_color(img_path)
_, im_h, im_w = img.shape
dets = tvm.relay.testing.yolo_detection.fill_network_boxes((netw, neth), (im_w, im_h), thresh,
                                                      1, tvm_out)
last_layer = net.layers[net.n - 1]
tvm.relay.testing.yolo_detection.do_nms_sort(dets, last_layer.classes, nms_thresh)

coco_name = 'coco.names'
coco_url = REPO_URL + 'data/' + coco_name + '?raw=true'
font_name = 'arial.ttf'
font_url = REPO_URL + 'data/' + font_name + '?raw=true'
coco_path = download_testdata(coco_url, coco_name, module='data')
font_path = download_testdata(font_url, font_name, module='data')

with open(coco_path) as f:
    content = f.readlines()

names = [x.strip() for x in content]

tvm.relay.testing.yolo_detection.draw_detections(font_path, img, dets, thresh, names, last_layer.classes)
plt.imshow(img.transpose(1, 2, 0))


Running the test image...
File /workspace/.tvm_test_data/data/coco.names exists, skip.
File /workspace/.tvm_test_data/data/arial.ttf exists, skip.

Total running time of the script: ( 0 minutes 34.304 seconds)

Gallery generated by Sphinx-Gallery