Commit
•
4f44aaf
1
Parent(s):
f129cdd
updates
Browse files- utils/utils.py +22 -7
utils/utils.py
CHANGED
@@ -471,6 +471,7 @@ def non_max_suppression(prediction, conf_thres=0.1, iou_thres=0.6, multi_label=T
|
|
471 |
# Settings
|
472 |
merge = True # merge for best mAP
|
473 |
min_wh, max_wh = 2, 4096 # (pixels) minimum and maximum box width and height
|
|
|
474 |
time_limit = 10.0 # seconds to quit after
|
475 |
|
476 |
t = time.time()
|
@@ -520,6 +521,8 @@ def non_max_suppression(prediction, conf_thres=0.1, iou_thres=0.6, multi_label=T
|
|
520 |
c = x[:, 5] * 0 if agnostic else x[:, 5] # classes
|
521 |
boxes, scores = x[:, :4].clone() + c.view(-1, 1) * max_wh, x[:, 4] # boxes (offset by class), scores
|
522 |
i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
|
|
|
|
|
523 |
if merge and (1 < n < 3E3): # Merge NMS (boxes merged using weighted mean)
|
524 |
try: # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
|
525 |
iou = box_iou(boxes[i], boxes) > iou_thres # iou matrix
|
@@ -975,15 +978,27 @@ def plot_targets_txt(): # from utils.utils import *; plot_targets_txt()
|
|
975 |
|
976 |
def plot_study_txt(f='study.txt', x=None): # from utils.utils import *; plot_study_txt()
|
977 |
# Plot study.txt generated by test.py
|
978 |
-
y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
|
979 |
-
x = np.arange(y.shape[1]) if x is None else np.array(x)
|
980 |
-
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_inference (ms/img)', 't_NMS (ms/img)', 't_total (ms/img)']
|
981 |
fig, ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)
|
982 |
ax = ax.ravel()
|
983 |
-
|
984 |
-
|
985 |
-
|
986 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
987 |
|
988 |
|
989 |
def plot_labels(labels):
|
|
|
471 |
# Settings
|
472 |
merge = True # merge for best mAP
|
473 |
min_wh, max_wh = 2, 4096 # (pixels) minimum and maximum box width and height
|
474 |
+
max_det = 300 # maximum number of detections per image
|
475 |
time_limit = 10.0 # seconds to quit after
|
476 |
|
477 |
t = time.time()
|
|
|
521 |
c = x[:, 5] * 0 if agnostic else x[:, 5] # classes
|
522 |
boxes, scores = x[:, :4].clone() + c.view(-1, 1) * max_wh, x[:, 4] # boxes (offset by class), scores
|
523 |
i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
|
524 |
+
if i.shape[0] > max_det: # limit detections
|
525 |
+
i = i[:max_det]
|
526 |
if merge and (1 < n < 3E3): # Merge NMS (boxes merged using weighted mean)
|
527 |
try: # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
|
528 |
iou = box_iou(boxes[i], boxes) > iou_thres # iou matrix
|
|
|
978 |
|
979 |
def plot_study_txt(f='study.txt', x=None): # from utils.utils import *; plot_study_txt()
|
980 |
# Plot study.txt generated by test.py
|
|
|
|
|
|
|
981 |
fig, ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)
|
982 |
ax = ax.ravel()
|
983 |
+
|
984 |
+
for f in glob.glob('study*.txt'):
|
985 |
+
y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
|
986 |
+
x = np.arange(y.shape[1]) if x is None else np.array(x)
|
987 |
+
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_inference (ms/img)', 't_NMS (ms/img)', 't_total (ms/img)']
|
988 |
+
for i in range(7):
|
989 |
+
ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
|
990 |
+
ax[i].set_title(s[i])
|
991 |
+
|
992 |
+
j = y[3].argmax() + 1
|
993 |
+
ax[7].plot(y[6, :j], y[3, :j] * 1E2, '.-', linewidth=2, markersize=8, label=Path(f).stem)
|
994 |
+
ax[7].plot(1E3 / np.array([209, 140, 97, 58, 35, 18]), [33.5, 39.1, 42.5, 45.9, 49., 50.5],
|
995 |
+
'.-', linewidth=2, markersize=8, label='EfficientDet')
|
996 |
+
ax[7].set_xlabel('Latency (ms)')
|
997 |
+
ax[7].set_ylabel('COCO AP val')
|
998 |
+
|
999 |
+
ax[7].legend()
|
1000 |
+
ax[7].set_xlim(0)
|
1001 |
+
plt.savefig(f.replace('.txt', '.png'), dpi=200)
|
1002 |
|
1003 |
|
1004 |
def plot_labels(labels):
|