def validate(net, device, val_data, batches_per_epoch):
"""
Run validation.
:param net: Network
:param device: Torch device
:param val_data: Validation Dataset
:param batches_per_epoch: Number of batches to run
:return: Successes, Failures and Losses
"""
net.eval()
results = {
'correct': 0,
'failed': 0,
'loss': 0,
'losses': {
}
}
ld = len(val_data)
with torch.no_grad():
batch_idx = 0
while batch_idx < batches_per_epoch:
for x, y, didx, rot, zoom_factor in val_data:
batch_idx += 1
if batches_per_epoch is not None and batch_idx >= batches_per_epoch:
break
xc = x.to(device)
yc = [yy.to(device) for yy in y]
lossd = net.compute_loss(xc, yc)
loss = lossd['loss']
results['loss'] += loss.item()/ld
for ln, l in lossd['losses'].items():
if ln not in results['losses']:
results['losses'][ln] = 0
results['losses'][ln] += l.item()/ld
q_out, ang_out, w_out = post_process_output(lossd['pred']['pos'], lossd['pred']['cos'],
lossd['pred']['sin'], lossd['pred']['width'])
s = evaluation.calculate_iou_match(q_out, ang_out,
val_data.dataset.get_gtbb(didx, rot, zoom_factor),
no_grasps=1,
grasp_width=w_out,
)
if s:
results['correct'] += 1
else:
results['failed'] += 1
return results
with torch.no_grad():
for idx, (x, y, didx, rot, zoom) in enumerate(test_data):
logging.info('Processing {}/{}'.format(idx + 1, len(test_data)))
xc = x.to(device)
yc = [yi.to(device) for yi in y]
lossd = net.compute_loss(xc, yc)
q_img, ang_img, width_img = post_process_output(
lossd['pred']['pos'], lossd['pred']['cos'],
lossd['pred']['sin'], lossd['pred']['width'])
if args.iou_eval:
s = evaluation.calculate_iou_match(
q_img,
ang_img,
test_data.dataset.get_gtbb(didx, rot, zoom),
no_grasps=args.n_grasps,
grasp_width=width_img,
)
if s:
results['correct'] += 1
else:
results['failed'] += 1
if args.jacquard_output:
grasps = grasp.detect_grasps(q_img,
ang_img,
width_img=width_img,
no_grasps=1)
with open(jo_fn, 'a') as f:
for g in grasps: