148 lines
5.2 KiB
Python
148 lines
5.2 KiB
Python
import torch
|
|
import numpy as np
|
|
import torchvision
|
|
|
|
def xywh2xyxy(x):
|
|
"""Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right"""
|
|
y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
|
|
y[..., 0] = x[..., 0] - x[..., 2] / 2 # top left x
|
|
y[..., 1] = x[..., 1] - x[..., 3] / 2 # top left y
|
|
y[..., 2] = x[..., 0] + x[..., 2] / 2 # bottom right x
|
|
y[..., 3] = x[..., 1] + x[..., 3] / 2 # bottom right y
|
|
return y
|
|
|
|
def box_iou(box1, box2):
|
|
"""
|
|
Return intersection-over-union (Jaccard index) of boxes.
|
|
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
|
|
box1: [N, 4]
|
|
box2: [M, 4]
|
|
Returns: [N, M]
|
|
"""
|
|
def box_area(box):
|
|
return (box[:, 2] - box[:, 0]) * (box[:, 3] - box[:, 1])
|
|
|
|
area1 = box_area(box1)
|
|
area2 = box_area(box2)
|
|
|
|
lt = torch.max(box1[:, None, :2], box2[:, :2]) # [N,M,2]
|
|
rb = torch.min(box1[:, None, 2:], box2[:, 2:]) # [N,M,2]
|
|
|
|
wh = (rb - lt).clamp(min=0) # [N,M,2]
|
|
inter = wh[:, :, 0] * wh[:, :, 1] # [N,M]
|
|
|
|
union = area1[:, None] + area2 - inter
|
|
return inter / (union + 1e-6)
|
|
|
|
def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, max_det=300):
|
|
"""
|
|
Non-Maximum Suppression (NMS) on inference results
|
|
prediction: [Batch, 84, 8400] (for YOLOv8/11)
|
|
"""
|
|
# [Batch, 84, Anchors] -> [Batch, Anchors, 84]
|
|
prediction = prediction.transpose(1, 2)
|
|
|
|
bs = prediction.shape[0]
|
|
nc = prediction.shape[2] - 4
|
|
xc = prediction[..., 4:].max(-1)[0] > conf_thres
|
|
|
|
output = [torch.zeros((0, 6), device=prediction.device)] * bs
|
|
|
|
for xi, x in enumerate(prediction):
|
|
x = x[xc[xi]]
|
|
if not x.shape[0]:
|
|
continue
|
|
|
|
# Box decoding
|
|
box = xywh2xyxy(x[:, :4])
|
|
|
|
# Confidence and Class
|
|
conf, j = x[:, 4:].max(1, keepdim=True)
|
|
x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
|
|
|
|
n = x.shape[0]
|
|
if not n:
|
|
continue
|
|
elif n > max_det:
|
|
x = x[x[:, 4].argsort(descending=True)[:max_det]]
|
|
|
|
# Batched NMS
|
|
c = x[:, 5:6] * 7680
|
|
boxes, scores = x[:, :4] + c, x[:, 4]
|
|
i = torchvision.ops.nms(boxes, scores, iou_thres)
|
|
output[xi] = x[i]
|
|
|
|
return output
|
|
|
|
def compute_ap(recall, precision):
|
|
""" Compute the average precision, given the recall and precision curves """
|
|
# Append sentinel values to beginning and end
|
|
mrec = np.concatenate(([0.0], recall, [1.0]))
|
|
mpre = np.concatenate(([1.0], precision, [0.0]))
|
|
|
|
# Compute the precision envelope
|
|
mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
|
|
|
|
# Integrate area under curve
|
|
method = 'interp' # methods: 'continuous', 'interp'
|
|
if method == 'interp':
|
|
x = np.linspace(0, 1, 101) # 101-point interp (COCO)
|
|
ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate
|
|
else: # 'continuous'
|
|
i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes
|
|
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve
|
|
|
|
return ap, mpre, mrec
|
|
|
|
def smooth(y, f=0.05):
|
|
"""Box filter of fraction f"""
|
|
nf = round(len(y) * f * 2) // 2 + 1 # number of filter elements (must be odd)
|
|
p = np.ones(nf // 2) # ones padding
|
|
yp = np.concatenate((p * y[0], y, p * y[-1]), 0) # y padded
|
|
return np.convolve(yp, np.ones(nf) / nf, mode='valid') # y-smoothed
|
|
|
|
def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16):
|
|
""" Compute the average precision, given the recall and precision curves. """
|
|
# Sort by objectness
|
|
i = np.argsort(-conf)
|
|
tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
|
|
|
|
# Find unique classes
|
|
unique_classes, nt = np.unique(target_cls, return_counts=True)
|
|
nc = unique_classes.shape[0] # number of classes, number of detections
|
|
|
|
# Create Precision-Recall curve and compute AP for each class
|
|
px, py = np.linspace(0, 1, 1000), [] # for plotting
|
|
ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
|
|
for ci, c in enumerate(unique_classes):
|
|
i = pred_cls == c
|
|
n_l = (target_cls == c).sum() # number of labels
|
|
n_p = i.sum() # number of predictions
|
|
|
|
if n_p == 0 or n_l == 0:
|
|
continue
|
|
|
|
# Accumulate FPs and TPs
|
|
fpc = (1 - tp[i]).cumsum(0)
|
|
tpc = tp[i].cumsum(0)
|
|
|
|
# Recall
|
|
recall = tpc / (n_l + eps) # recall curve
|
|
r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases
|
|
|
|
# Precision
|
|
precision = tpc / (tpc + fpc) # precision curve
|
|
p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score
|
|
|
|
# AP from recall-precision curve
|
|
for j in range(tp.shape[1]):
|
|
ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
|
|
|
|
# Compute F1 (harmonic mean of precision and recall)
|
|
f1 = 2 * p * r / (p + r + eps)
|
|
i = smooth(f1.mean(0), 0.1).argmax() # max F1 index
|
|
p, r, f1 = p[:, i], r[:, i], f1[:, i]
|
|
tp = (r * nt).round().astype(int)
|
|
fp = (tp / (p + eps) - tp).astype(int)
|
|
|
|
return tp, fp, p, r, f1, ap, unique_classes.astype(int) |