YOLOv2使用笔记——grid cell、object confidence可视化
https://github.com/xueeinstein/darknet-vis
1.修改cfg
只需要在[net]中增加
visualization=12. 可视化
darkney.py
from ctypes import *
import os
THIS_DIR = os.path.dirname(os.path.realpath(__file__))
class BOX(Structure):
_fields_ = [("x", c_float), ("y", c_float), ("w", c_float), ("h", c_float)]
class IMAGE(Structure):
_fields_ = [("w", c_int), ("h", c_int), ("c", c_int), ("data",
POINTER(c_float))]
class METADATA(Structure):
_fields_ = [("classes", c_int), ("names", POINTER(c_char_p))]
class Darknet(object):
"""Wrapper for darknet APIs from dynamic lib"""
def __init__(self, cfg, meta, weights):
libdarknet_path = os.path.join(THIS_DIR, "..", "libdarknet.so")
lib = CDLL(libdarknet_path, RTLD_GLOBAL)
lib.network_width.argtypes = [c_void_p]
lib.network_width.restype = c_int
lib.network_height.argtypes = [c_void_p]
lib.network_height.restype = c_int
self.predict = lib.network_predict_p
self.predict.argtypes = [c_void_p, POINTER(c_float)]
self.predict.restype = POINTER(c_float)
self.make_boxes = lib.make_boxes
self.make_boxes.argtypes = [c_void_p]
self.make_boxes.restype = POINTER(BOX)
self.free_ptrs = lib.free_ptrs
self.free_ptrs.argtypes = [POINTER(c_void_p), c_int]
self.num_boxes = lib.num_boxes
self.num_boxes.argtypes = [c_void_p]
self.num_boxes.restype = c_int
self.make_probs = lib.make_probs
self.make_probs.argtypes = [c_void_p]
self.make_probs.restype = POINTER(POINTER(c_float))
self.reset_rnn = lib.reset_rnn
self.reset_rnn.argtypes = [c_void_p]
self.load_net = lib.load_network_p
self.load_net.argtypes = [c_char_p, c_char_p, c_int]
self.load_net.restype = c_void_p
self.free_image = lib.free_image
self.free_image.argtypes = [IMAGE]
self.letterbox_image = lib.letterbox_image
self.letterbox_image.argtypes = [IMAGE, c_int, c_int]
self.letterbox_image.restype = IMAGE
self.load_meta = lib.get_metadata
lib.get_metadata.argtypes = [c_char_p]
lib.get_metadata.restype = METADATA
self.load_image = lib.load_image_color
self.load_image.argtypes = [c_char_p, c_int, c_int]
self.load_image.restype = IMAGE
self.predict_image = lib.network_predict_image
self.predict_image.argtypes = [c_void_p, IMAGE]
self.predict_image.restype = POINTER(c_float)
self.network_detect = lib.network_detect
self.network_detect.argtypes = [
c_void_p, IMAGE, c_float, c_float, c_float,
POINTER(BOX),
POINTER(POINTER(c_float))
]
self.network_visual_detect = lib.network_visual_detect
self.network_visual_detect.argtypes = [
c_void_p, IMAGE, c_float, c_float, c_float,
POINTER(BOX),
POINTER(POINTER(c_float)),
POINTER(c_char_p),
POINTER(POINTER(IMAGE)), c_char_p
]
self.draw_detections = lib.draw_detections
self.draw_detections.argtypes = [
IMAGE, c_int, c_float,
POINTER(BOX),
POINTER(POINTER(c_float)),
POINTER(POINTER(c_float)),
POINTER(c_char_p),
POINTER(POINTER(IMAGE)), c_int, c_int
]
self.load_alphabet = lib.load_alphabet
self.load_alphabet.argtypes = []
self.load_alphabet.restype = POINTER(POINTER(IMAGE))
self.save_image = lib.save_image
self.save_image.argtypes = [IMAGE, c_char_p]
self._init_network(cfg, meta, weights)
def _init_network(self, cfg, meta, weights):
self.net = self.load_net(cfg, weights, 0)
self.meta_f = meta
self.meta = self.load_meta(meta)
def _get_names(self, meta):
names_file = None
with open(meta, 'w') as f:
for l in f.readlines():
if l.startswith("names"):
names_file = l.split('=')[1].strip()
break
if names_file is None:
raise Exception("Cannot find 'names' entry in {}".format(meta))
with open(names_file) as fn:
names = [i.strip() for i in fn.readlines()]
return names
def classify(self, im):
out = self.predict_image(self.net, im)
res = []
for i in range(self.meta.classes):
res.append((self.meta.names[i], out[i]))
res = sorted(res, key=lambda x: -x[1])
return res
def detect(self, image, thresh=.5, hier_thresh=.5, nms=.45, outfile=None):
im = self.load_image(image, 0, 0)
boxes = self.make_boxes(self.net)
probs = self.make_probs(self.net)
num = self.num_boxes(self.net)
self.network_detect(self.net, im, thresh, hier_thresh, nms, boxes,
probs)
# draw bounding boxes
if outfile is not None:
masks = POINTER(POINTER(c_float))()
alphabet = self.load_alphabet()
names = self.meta.names
classes = self.meta.classes
self.draw_detections(im, num, thresh, boxes, probs, masks, names,
alphabet, classes, 1)
self.save_image(im, outfile)
res = []
for j in range(num):
for i in range(self.meta.classes):
if probs[j][i] > 0:
res.append((self.meta.names[i], probs[j][i],
(boxes[j].x, boxes[j].y, boxes[j].w,
boxes[j].h)))
res = sorted(res, key=lambda x: -x[1])
self.free_image(im)
self.free_ptrs(cast(probs, POINTER(c_void_p)), num)
return res
def detect_v(self, image, outfile, thresh=.5, hier_thresh=.5, nms=.45):
im = self.load_image(image, 0, 0)
boxes = self.make_boxes(self.net)
probs = self.make_probs(self.net)
num = self.num_boxes(self.net)
alphabet = self.load_alphabet()
names = self.meta.names
self.network_visual_detect(self.net, im, thresh, hier_thresh, nms,
boxes, probs, names, alphabet, outfile)
res = []
for j in range(num):
for i in range(self.meta.classes):
if probs[j][i] > 0:
res.append((self.meta.names[i], probs[j][i],
(boxes[j].x, boxes[j].y, boxes[j].w,
boxes[j].h)))
res = sorted(res, key=lambda x: -x[1])
self.free_image(im)
self.free_ptrs(cast(probs, POINTER(c_void_p)), num)
return res
if __name__ == "__main__":
cfg = os.path.join(THIS_DIR, "..", "cfg", "yolo.cfg")
meta = os.path.join(THIS_DIR, "..", "cfg", "coco.data")
weights = os.path.join(THIS_DIR, "..", "yolo.weights")
img = os.path.join(THIS_DIR, "..", "data", "dog.jpg")
yolo = Darknet(cfg, meta, weights)
#res = yolo.detect(img, outfile="yolo_det")
res = yolo.detect_v(img, outfile="yolo_det")
print(res)python python/darknet.py
通过grid cell网格化，每个网络各自负责预测一些bounding box，分别与ground truth计算iou，大于等于0即正样本，小于0作为负样本训练。
计算confidence，每个iou大于0的bounding box进分类网络，得到class probalility，再乘以iou。
通过可视化理解原理
https://blog.csdn.net/zfq740695564/article/details/79754578