def writelog(**kwargs):
global best
global bestn
opt = kwargs['opt']
iepoch = kwargs['iepoch']
nepoch = opt['nepoch']
ib = kwargs['idata']
nb = kwargs['ndata'] // opt['batch_size']
out = kwargs['out']
net = kwargs['net']
d = kwargs['data']
meter = kwargs['meter']
optim = kwargs['optim']
print('[' + str(datetime.now()) + '][%d/%d,%d/%d]' %
(iepoch, nepoch, ib, nb) + 'training:' + str(kwargs['istraining']))
if not 'log_tmp' in opt.keys():
opt['log_tmp'] = opt['log'] + os.sep + opt['net'] + '_' + opt[
'config'] + '_' + opt['mode'] + '_' + str(datetime.now()).replace(
' ', '-').replace(':', '-')
os.mkdir(opt['log_tmp'])
with open(opt['log_tmp'] + os.sep + 'options.json', 'w') as f:
json.dump(opt, f, cls=NpEncoder)
nparam = 0
with open(opt['log_tmp'] + os.sep + 'net.txt', 'w') as logtxt:
print(str(kwargs['net']), file=logtxt)
for p in parameters(kwargs['net']):
nparam += torch.numel(p)
print('nparam:%d' % nparam, file=logtxt)
with open(opt['log_tmp'] + os.sep + 'log.txt', 'a') as logtxt:
print('[' + str(datetime.now()) + '][%d/%d,%d/%d]' %
(iepoch, nepoch, ib, nb) + 'training:' +
str(kwargs['istraining']),
file=logtxt)
print(json.dumps(meter, cls=NpEncoder), file=logtxt)
if kwargs['istraining']:
tfb_dir = os.path.join(opt['log_tmp'], 'tfb', 'train')
else:
tfb_dir = os.path.join(opt['log_tmp'], 'tfb', 'test')
write_tfb(tfb_dir, meter, ib + nb * iepoch, nb, optim)
if not kwargs['istraining'] and ib >= nb - 1:
if meter['box'].overall_meter.avg < best[-1]:
fn = bestn[-1]
if fn:
os.remove(opt['log_tmp'] + os.sep + fn)
fn = 'net_' + str(datetime.now()).replace(' ', '-').replace(
':', '-') + '.pth'
best[-1] = meter['box'].overall_meter.avg
bestn[-1] = fn
torch.save(net.state_dict(), opt['log_tmp'] + os.sep + fn)
idx = np.argsort(best)
best = best[idx]
bestn = [bestn[x] for x in idx.tolist()]
bestdict = dict(zip(bestn, best.tolist()))
print(best)
print(bestn)
print(bestdict)
with open(opt['log_tmp'] + os.sep + 'best.json', 'w') as f:
json.dump(bestdict, f)
def writelog(**kwargs):
global best;
global bestn;
opt = kwargs['opt'];
iepoch = kwargs['iepoch'];
nepoch = opt['nepoch'];
ib = kwargs['idata'];
nb = kwargs['ndata'] // opt['batch_size'];
out = kwargs['out'];
net = kwargs['net'];
d = kwargs['data'];
meter = kwargs['meter'];
optim = kwargs['optim'];
print('['+str(datetime.now())+'][%d/%d,%d/%d]'%(iepoch,nepoch,ib,nb)+'training:'+str(kwargs['istraining']));
if not 'log_tmp' in opt.keys():
opt['log_tmp'] = opt['log']+os.sep+opt['net']+'_'+opt['config']+'_'+opt['mode']+'_'+str(datetime.now()).replace(' ','-').replace(':','-');
os.mkdir(opt['log_tmp']);
with open(opt['log_tmp']+os.sep+'options.json','w') as f:
json.dump(opt,f,cls=NpEncoder);
nparam = 0;
with open(opt['log_tmp']+os.sep+'net.txt','w') as logtxt:
print(str(kwargs['net']),file=logtxt);
for p in parameters(kwargs['net']):
nparam += torch.numel(p);
print('nparam:%d'%nparam,file=logtxt);
with open(opt['log_tmp']+os.sep+'log.txt','a') as logtxt:
print('['+str(datetime.now())+'][%d/%d,%d/%d]'%(iepoch,nepoch,ib,nb)+'training:'+str(kwargs['istraining']),file=logtxt);
print(json.dumps(meter,cls=NpEncoder),file=logtxt);
if kwargs['istraining']:
tfb_dir = os.path.join(opt['log_tmp'],'tfb','train');
else:
tfb_dir = os.path.join(opt['log_tmp'],'tfb','test');
write_tfb(tfb_dir,meter,ib+nb*iepoch,nb,optim);
if not kwargs['istraining'] and ib >= nb-1:
if meter['cd'].overall_meter.avg < best[-1]:
fn = bestn[-1];
if fn:
os.remove(opt['log_tmp']+os.sep+fn);
fn = 'net_'+str(datetime.now()).replace(' ','-').replace(':','-')+'.pth';
best[-1] = meter['cd'].overall_meter.avg;
bestn[-1] = fn;
torch.save(net.state_dict(),opt['log_tmp']+os.sep+fn);
idx = np.argsort(best);
best = best[idx];
bestn = [bestn[x] for x in idx.tolist()];
bestdict = dict(zip(bestn, best.tolist()));
print(best);
print(bestn);
print(bestdict);
with open(opt['log_tmp']+os.sep+'best.json','w') as f:
json.dump(bestdict,f);
if opt['ply'] and not kwargs['istraining']:
ply_path = opt['log_tmp']+os.sep+'ply';
if not os.path.exists(ply_path):
os.mkdir(ply_path);
img = d[0].data.cpu().numpy();
box2d_src = d[1].data.cpu().numpy();
box3d_src = d[2].data.cpu().numpy();
box2d_tgt = d[3].data.cpu().numpy();
box3d_tgt = d[4].data.cpu().numpy();
rs = d[5].data.cpu().numpy();
gts = d[6].data.cpu().numpy();
ys = out['y'].data.cpu().numpy();
cat = d[-1];
for i in range(img.shape[0]):
fig = plt.figure();
r = rs[i,...];
y = ys[i,...];
y *= np.pi;
y[1] *= 2;
coord = np.concatenate([r.reshape(-1,1),y.reshape(-1,2)],axis=1);
c3dir = sph2car(coord.reshape(1,-1));
c3d = np.zeros([2,3],dtype=np.float32);
c3d[1,:] = mv(np.mean(box3d_src[i,:,:3],axis=0,keepdims=True))[:,:3];
c3d[0,:] = c3d[1,:] + c3dir;
c3d = mv_inv(c3d);
c2d = proj(mv(c3d));
#
gt = gts[i,...];
gt *= np.pi;
gt[1] *= 2;
coord2 = np.concatenate([r.reshape(-1,1),gt.reshape(-1,2)],axis=1);
c3dir2 = sph2car(coord2.reshape(1,-1));
c3d2 = np.zeros([2,3],dtype=np.float32);
c3d2[1,:] = mv(np.mean(box3d_src[i,:,:3],axis=0,keepdims=True))[:,:3];
c3d2[0,:] = c3d2[1,:] + c3dir2;
c3d2 = mv_inv(c3d2);
#
ax = fig.add_subplot(121);
ax.imshow(img[i,...]);
ax.set_aspect('equal');
ax.scatter(box2d_src[i,0:4,0],box2d_src[i,0:4,1],color='b',marker='*');
ax.scatter(box2d_src[i,4:8,0],box2d_src[i,4:8,1],color='c',marker='*');
ax.scatter(box2d_tgt[i,0:4,0],box2d_tgt[i,0:4,1],color='k',marker='x');
ax.scatter(box2d_tgt[i,4:8,0],box2d_tgt[i,4:8,1],color='r',marker='x');
ax.scatter(c2d[0,0],c2d[0,1],color='b',marker='o');
ax.scatter(c2d[1,0],c2d[1,1],color='r',marker='o');
ax = fig.add_subplot(122,projection='3d');
#ax.set_aspect('equal');
ax.scatter(box3d_src[i,0:4,0],box3d_src[i,0:4,1],box3d_src[i,0:4,2],color='b',marker='*');
ax.scatter(box3d_src[i,4:8,0],box3d_src[i,4:8,1],box3d_src[i,4:8,2],color='c',marker='*');
ax.scatter(box3d_tgt[i,0:4,0],box3d_tgt[i,0:4,1],box3d_tgt[i,0:4,2],color='k',marker='x');
ax.scatter(box3d_tgt[i,4:8,0],box3d_tgt[i,4:8,1],box3d_tgt[i,4:8,2],color='r',marker='x');
ax.plot(c3d[:,0],c3d[:,1],c3d[:,2],color='k');
ax.plot(c3d2[:,0],c3d2[:,1],c3d2[:,2],color='r');
cs = np.mean(box3d_src[i,:,:3],axis=0);
ax.scatter(cs[0],cs[1],cs[2],color='r',marker='o');
ct = np.mean(box3d_tgt[i,:,:3],axis=0);
ax.scatter(ct[0],ct[1],ct[2],color='b',marker='o');
plt.savefig(ply_path+os.sep+'_%04d_%03d_%s.png'%(ib,i,cat[i]));
plt.close(fig);