Skip to content

P
python3-imagej-tiff
Commit 0452a446 authored by Oleg Dzhimiev's avatar Oleg Dzhimiev
Browse files
Options

1. packing type 2 

2. epochs, batches changes
parent 45a473f0
Hide whitespace changes
Inline Side-by-side
Showing with 276 additions and 110 deletions
pack_tile.py
View file @ 0452a446
......@@ -6,7 +6,8 @@ __email__ = "oleg@elphel.com"
import numpy as np
# pack from 9x9x4 to 25x1
# pack from 9x9x4 to 100x1
# type 1, all the same
def pack_tile_type1(tile):
out = np.empty(100)
......@@ -125,13 +126,143 @@ def pack_tile_type1(tile):
return out
# pack from 9x9x4 to 104x1
# type 1, all the same
def pack_tile_type2(tile):
out = np.empty(104)
# pack diagm-pair (not tested)
l = np.ravel(tile[:,:,0])
out[ 0] = 1.0*l[ 0]+1.0*l[ 1]+1.0*l[ 2]+1.0*l[ 9]+1.0*l[10]+1.0*l[18]
out[ 1] = 1.0*l[ 3]+1.0*l[11]+1.0*l[19]+1.0*l[27]
out[ 2] = 1.0*l[ 4]+1.0*l[12]+1.0*l[20]+1.0*l[28]+1.0*l[36]
out[ 3] = 1.0*l[ 5]+1.0*l[ 6]+1.0*l[14]
out[ 4] = 1.0*l[ 7]+1.0*l[15]+1.0*l[23]
out[ 5] = 1.0*l[ 8]+1.0*l[16]+1.0*l[24]
out[ 6] = 1.0*l[13]+1.0*l[21]+1.0*l[29]+1.0*l[37]
out[ 7] = 1.0*l[17]+1.0*l[25]+1.0*l[33]
out[ 8] = 1.0*l[22]+1.0*l[30]+1.0*l[38]
out[ 9] = 1.0*l[26]+1.0*l[34]+1.0*l[35]
out[10] = 1.0*l[31]
out[11] = 1.0*l[32]
out[12] = 1.0*l[39]
out[13] = 1.0*l[40]
out[14] = 1.0*l[41]
out[15] = 1.0*l[42]+1.0*l[50]+1.0*l[58]
out[16] = 1.0*l[43]+1.0*l[51]+1.0*l[59]+1.0*l[67]
out[17] = 1.0*l[44]+1.0*l[52]+1.0*l[60]+1.0*l[68]+1.0*l[76]
out[18] = 1.0*l[45]+1.0*l[46]+1.0*l[54]
out[19] = 1.0*l[47]+1.0*l[55]+1.0*l[63]
out[20] = 1.0*l[48]
out[21] = 1.0*l[49]
out[22] = 1.0*l[53]+1.0*l[61]+1.0*l[69]+1.0*l[77]
out[23] = 1.0*l[56]+1.0*l[64]+1.0*l[72]
out[24] = 1.0*l[57]+1.0*l[65]+1.0*l[73]
out[25] = 1.0*l[62]+1.0*l[70]+1.0*l[71]+1.0*l[78]+1.0*l[79]+1.0*l[80]
out[26] = 1.0*l[66]+1.0*l[74]+1.0*l[75]
# pack diago-pair (not tested)
l = np.ravel(tile[:,:,1])
out[27] = 1.0*l[ 0]+1.0*l[10]+1.0*l[20]
out[28] = 1.0*l[ 1]+1.0*l[11]+1.0*l[21]
out[29] = 1.0*l[ 2]+1.0*l[ 3]+1.0*l[12]
out[30] = 1.0*l[ 4]+1.0*l[14]+1.0*l[24]+1.0*l[34]+1.0*l[44]
out[31] = 1.0*l[ 5]+1.0*l[15]+1.0*l[25]+1.0*l[35]
out[32] = 1.0*l[ 6]+1.0*l[ 7]+1.0*l[ 8]+1.0*l[16]+1.0*l[17]+1.0*l[26]
out[33] = 1.0*l[ 9]+1.0*l[19]+1.0*l[29]
out[34] = 1.0*l[13]+1.0*l[23]+1.0*l[43]
out[35] = 1.0*l[18]+1.0*l[27]+1.0*l[28]
out[36] = 1.0*l[22]+1.0*l[32]+1.0*l[42]
out[37] = 1.0*l[30]
out[38] = 1.0*l[31]
out[39] = 1.0*l[36]+1.0*l[46]+1.0*l[56]+1.0*l[66]+1.0*l[76]
out[40] = 1.0*l[37]+1.0*l[47]+1.0*l[57]+1.0*l[67]
out[41] = 1.0*l[38]+1.0*l[48]+1.0*l[58]
out[42] = 1.0*l[39]
out[43] = 1.0*l[40]
out[44] = 1.0*l[41]
out[45] = 1.0*l[45]+1.0*l[55]+1.0*l[65]+1.0*l[75]
out[46] = 1.0*l[49]
out[47] = 1.0*l[50]
out[48] = 1.0*l[51]+1.0*l[61]+1.0*l[71]
out[49] = 1.0*l[52]+1.0*l[53]+1.0*l[62]
out[50] = 1.0*l[54]+1.0*l[63]+1.0*l[64]+1.0*l[72]+1.0*l[73]+1.0*l[74]
out[51] = 1.0*l[59]+1.0*l[69]+1.0*l[79]
out[52] = 1.0*l[60]+1.0*l[70]+1.0*l[80]
out[53] = 1.0*l[68]+1.0*l[77]+1.0*l[78]
# pack hor-pairs
l = np.ravel(tile[:,:,2])
out[54] = 1.0*l[ 0]+1.0*l[ 1]+1.0*l[ 9]+1.0*l[10]+1.0*l[18]+1.0*l[27]+1.0*l[36]+1.0*l[45]+1.0*l[54]+1.0*l[63]+1.0*l[64]+1.0*l[72]+1.0*l[73]
out[55] = 1.0*l[ 2]+1.0*l[11]+1.0*l[20]
out[56] = 1.0*l[ 3]+1.0*l[12]+1.0*l[21]
out[57] = 1.0*l[ 4]+1.0*l[13]+1.0*l[22]
out[58] = 1.0*l[ 5]+1.0*l[14]+1.0*l[23]
out[59] = 1.0*l[ 6]+1.0*l[15]+1.0*l[24]
out[60] = 1.0*l[ 7]+1.0*l[ 8]+1.0*l[16]+1.0*l[17]+1.0*l[26]+1.0*l[35]+1.0*l[44]+1.0*l[53]+1.0*l[62]+1.0*l[70]+1.0*l[71]+1.0*l[79]+1.0*l[80]
out[61] = 1.0*l[19]+1.0*l[28]+1.0*l[37]+1.0*l[46]+1.0*l[55]
out[62] = 1.0*l[25]+1.0*l[34]+1.0*l[43]+1.0*l[52]+1.0*l[61]
out[63] = 1.0*l[29]+1.0*l[38]+1.0*l[47]
out[64] = 1.0*l[30]
out[65] = 1.0*l[31]
out[66] = 1.0*l[32]
out[67] = 1.0*l[33]+1.0*l[42]+1.0*l[51]
out[68] = 1.0*l[39]
out[69] = 1.0*l[40]
out[70] = 1.0*l[41]
out[71] = 1.0*l[48]
out[72] = 1.0*l[49]
out[73] = 1.0*l[50]
out[74] = 1.0*l[56]+1.0*l[65]+1.0*l[74]
out[75] = 1.0*l[57]+1.0*l[66]+1.0*l[75]
out[76] = 1.0*l[58]+1.0*l[67]+1.0*l[76]
out[77] = 1.0*l[59]+1.0*l[68]+1.0*l[77]
out[78] = 1.0*l[60]+1.0*l[69]+1.0*l[78]
# pack vert-pairs
l = np.ravel(tile[:,:,3])
out[79] = 1.0*l[ 0]+1.0*l[ 1]+1.0*l[ 2]+1.0*l[ 3]+1.0*l[ 4]+1.0*l[ 5]+1.0*l[ 6]+1.0*l[ 7]+1.0*l[ 8]+1.0*l[ 9]+1.0*l[10]+1.0*l[11]+1.0*l[16]+1.0*l[17]
out[80] = 1.0*l[11]+1.0*l[12]+1.0*l[13]+1.0*l[14]+1.0*l[15]
out[81] = 1.0*l[18]+1.0*l[19]+1.0*l[20]
out[82] = 1.0*l[21]+1.0*l[22]+1.0*l[23]
out[83] = 1.0*l[24]+1.0*l[25]+1.0*l[26]
out[84] = 1.0*l[27]+1.0*l[28]+1.0*l[29]
out[85] = 1.0*l[30]
out[86] = 1.0*l[31]
out[87] = 1.0*l[32]
out[88] = 1.0*l[33]+1.0*l[34]+1.0*l[35]
out[89] = 1.0*l[36]+1.0*l[37]+1.0*l[38]
out[90] = 1.0*l[39]
out[91] = 1.0*l[40]
out[92] = 1.0*l[41]
out[93] = 1.0*l[42]+1.0*l[43]+1.0*l[44]
out[94] = 1.0*l[45]+1.0*l[46]+1.0*l[47]
out[95] = 1.0*l[48]
out[96] = 1.0*l[49]
out[97] = 1.0*l[50]
out[98] = 1.0*l[51]+1.0*l[52]+1.0*l[53]
out[99] = 1.0*l[54]+1.0*l[55]+1.0*l[56]
out[100] = 1.0*l[57]+1.0*l[58]+1.0*l[59]
out[101] = 1.0*l[60]+1.0*l[61]+1.0*l[62]
out[102] = 1.0*l[63]+1.0*l[64]+1.0*l[70]+1.0*l[71]+1.0*l[72]+1.0*l[73]+1.0*l[74]+1.0*l[75]+1.0*l[76]+1.0*l[77]+1.0*l[78]+1.0*l[79]+1.0*l[80]
out[103] = 1.0*l[65]+1.0*l[66]+1.0*l[67]+1.0*l[68]+1.0*l[69]
return out
# pack single
def pack_tile(tile):
return pack_tile_type1(tile)
# pack all tiles
def pack(tiles):
output = np.array([[pack_tile(tiles[i,j]) for j in range(tiles.shape[1])] for i in range(tiles.shape[0])])
def pack(tiles,ptype=1):
if ptype==1:
pack_func = pack_tile_type1
elif ptype==2:
pack_func = pack_tile_type2
output = np.array([[pack_func(tiles[i,j]) for j in range(tiles.shape[1])] for i in range(tiles.shape[0])])
return output
......
test_nn_feed.py
View file @ 0452a446
......@@ -46,6 +46,7 @@ def print_time():
VALUES_LAYER_NAME = 'other'
LAYERS_OF_INTEREST = ['diagm-pair', 'diago-pair', 'hor-pairs', 'vert-pairs']
RADIUS = 1
TILE_PACKING_TYPE = 1
DEBUG_PLT_LOSS = True
# If false - will not pack or rescal
......@@ -91,23 +92,27 @@ def lrelu(x):
def network(input):
fc1 = slim.fully_connected(input,512,activation_fn=lrelu,scope='g_fc1')
fc2 = slim.fully_connected(fc1, 2,activation_fn=lrelu,scope='g_fc2')
return fc2
fc1 = slim.fully_connected(input,1024,activation_fn=lrelu,scope='g_fc1')
#fc2 = slim.fully_connected(fc1, 2,activation_fn=lrelu,scope='g_fc2')
#return fc2
#fc2 = slim.fully_connected(fc1, 1024,activation_fn=lrelu,scope='g_fc2')
#fc3 = slim.fully_connected(fc2, 512,activation_fn=lrelu,scope='g_fc3')
#fc4 = slim.fully_connected(fc3, 8,activation_fn=lrelu,scope='g_fc4')
#fc5 = slim.fully_connected(fc4, 4,activation_fn=lrelu,scope='g_fc5')
#fc6 = slim.fully_connected(fc5, 2,activation_fn=lrelu,scope='g_fc6')
fc2 = slim.fully_connected(fc1, 1024,activation_fn=lrelu,scope='g_fc2')
fc3 = slim.fully_connected(fc2, 512,activation_fn=lrelu,scope='g_fc3')
fc4 = slim.fully_connected(fc3, 8,activation_fn=lrelu,scope='g_fc4')
fc5 = slim.fully_connected(fc4, 4,activation_fn=lrelu,scope='g_fc5')
fc6 = slim.fully_connected(fc5, 2,activation_fn=lrelu,scope='g_fc6')
#return fc6
return fc6
sess = tf.Session()
in_tile = tf.placeholder(tf.float32,[None,101])
gt = tf.placeholder(tf.float32,[None,2])
if TILE_PACKING_TYPE==1:
in_tile = tf.placeholder(tf.float32,[None,101])
elif TILE_PACKING_TYPE==2:
in_tile = tf.placeholder(tf.float32,[None,105])
gt = tf.placeholder(tf.float32,[None,2])
#losses = tf.get_variable("losses", [None])
......@@ -135,23 +140,23 @@ cf_w_norm = tf.nn.softmax(cf_w)
#G_loss = tf.reduce_mean(tf.abs(tf.nn.softmax(out[:,1])*out[:,0]-cf_w_norm*gt[:,0]))
#G_loss = tf.reduce_mean(tf.squared_difference(out[:,0], gt[:,0]))
#G_loss = tf.reduce_mean(tf.abs(out[:,0]-gt[:,0]))
G_loss = tf.losses.mean_squared_error(gt[:,0],out[:,0],cf_w)
G_loss = tf.reduce_mean(tf.abs(out[:,0]-gt[:,0]))
#G_loss = tf.losses.mean_squared_error(gt[:,0],out[:,0],cf_w)
tf.summary.scalar('loss', G_loss)
tf.summary.scalar('prediction', out[0,0])
tf.summary.scalar('ground truth', gt[0,0])
#tf.summary.scalar('loss', G_loss)
#tf.summary.scalar('prediction', out[0,0])
#tf.summary.scalar('ground truth', gt[0,0])
t_vars=tf.trainable_variables()
lr=tf.placeholder(tf.float32)
G_opt=tf.train.AdamOptimizer(learning_rate=lr).minimize(G_loss,var_list=[var for var in t_vars if var.name.startswith('g_')])
G_opt=tf.train.AdamOptimizer(learning_rate=lr).minimize(G_loss)
saver=tf.train.Saver()
# ?!!!!!
merged = tf.summary.merge_all()
train_writer = tf.summary.FileWriter(result_dir + '/train', sess.graph)
test_writer = tf.summary.FileWriter(result_dir + '/test')
#merged = tf.summary.merge_all()
#train_writer = tf.summary.FileWriter(result_dir + '/train', sess.graph)
#test_writer = tf.summary.FileWriter(result_dir + '/test')
sess.run(tf.global_variables_initializer())
ckpt=tf.train.get_checkpoint_state(checkpoint_dir)
......@@ -185,132 +190,143 @@ if DEBUG_PLT_LOSS:
pass
training_tiles = np.array([])
training_values = np.array([])
# get epoch train data
for i in range(len(tlist)):
# RUN
# epoch is one image
print(bcolors.OKGREEN+"Opening "+tlist[i]+bcolors.ENDC)
tmp_tiff = ijt.imagej_tiff(tlist[i])
tmp_tiles = tmp_tiff.getstack(labels,shape_as_tiles=True)
tmp_vals = tmp_tiff.getvalues(label=VALUES_LAYER_NAME)
# might not need it because going to loop through anyway
if TILE_PACKING_TYPE==1:
packed_tiles = pile.pack(tmp_tiles)
elif TILE_PACKING_TYPE==2:
packed_tiles = pile.pack(tmp_tiles,TILE_PACKING_TYPE)
packed_tiles = np.dstack((packed_tiles,tmp_vals[:,:,0]))
for epoch in range(lastepoch,lastepoch+len(tlist)):
packed_tiles = np.reshape(packed_tiles,(-1,packed_tiles.shape[-1]))
values = np.reshape(tmp_vals[:,:,1:3],(-1,2))
print(bcolors.HEADER+"Epoch #"+str(epoch)+bcolors.ENDC)
packed_tiles_filtered = np.array([])
if os.path.isdir("result/%04d"%epoch):
continue
tlist_index = epoch - lastepoch
print("Unfiltered: "+str(packed_tiles.shape))
print(bcolors.OKGREEN+"Processing "+tlist[tlist_index]+bcolors.ENDC)
for j in range(packed_tiles.shape[0]):
tmp_tiff = ijt.imagej_tiff(tlist[tlist_index])
tmp_tiles = tmp_tiff.getstack(labels,shape_as_tiles=True)
tmp_vals = tmp_tiff.getvalues(label=VALUES_LAYER_NAME)
skip_tile = False
if np.isnan(np.sum(packed_tiles[j])):
skip_tile = True
if np.isnan(np.sum(values[j])):
skip_tile = True
# might not need it because going to loop through anyway
packed_tiles = pile.pack(tmp_tiles)
packed_tiles = np.dstack((packed_tiles,tmp_vals[:,:,0]))
if not skip_tile:
if len(packed_tiles_filtered)==0:
packed_tiles_filtered = np.array([packed_tiles[j]])
values_filtered = np.array([values[j]])
else:
packed_tiles_filtered = np.append(packed_tiles_filtered,[packed_tiles[j]],axis=0)
values_filtered = np.append(values_filtered,[values[j]],axis=0)
#if epoch > 2000:
# LR = 1e-5
print("NaN-filtered: "+str(packed_tiles_filtered.shape))
# so, here get the image, remove nans and run for 100x times
packed_tiles[np.isnan(packed_tiles)] = 0.0
tmp_vals[np.isnan(tmp_vals)] = 0.0
if i==0:
training_tiles = packed_tiles_filtered
training_values = values_filtered
else:
training_tiles = np.concatenate((training_tiles,packed_tiles_filtered),axis=0)
training_values = np.concatenate((training_values,values_filtered),axis=0)
#packed_tiles = packed_tiles[::,::]
values = tmp_vals
print("Training set shape: "+str(training_tiles.shape))
input_patch = np.reshape(packed_tiles,(-1,101))
gt_patch = np.reshape(values[:,:,1:3],(-1,2))
# RUN
# epoch is all available images
# batch is a number of non-zero tiles
g_loss = np.zeros(input_patch.shape[0])
g_loss = np.zeros(training_tiles.shape[0])
#for epoch in range(lastepoch,lastepoch+len(tlist)):
for epoch in range(lastepoch,500):
for i in range(100):
print(bcolors.HEADER+"Epoch #"+str(epoch)+bcolors.ENDC)
print(bcolors.OKBLUE+"Iteration "+str(i)+bcolors.ENDC)
if os.path.isdir("result/%04d"%epoch):
continue
st=time.time()
#if epoch > 2000:
# LR = 1e-5
skip_iteration = False
# so, here get the image, remove nans and run for 100x times
#packed_tiles[np.isnan(packed_tiles)] = 0.0
#tmp_vals[np.isnan(tmp_vals)] = 0.0
# if nan skip run!
if np.isnan(np.sum(gt_patch)):
print("GT has NaNs")
#skip_iteration = True
input_patch = training_tiles
gt_patch = training_values
if np.isnan(np.sum(input_patch)):
print("Patch has NaNs")
#skip_iteration = True
st=time.time()
if skip_iteration:
#print(bcolors.WARNING+"Found NaN, skipping iteration for tile "+str(i)+","+str(j)+bcolors.ENDC)
pass
else:
#run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
#run_metadata = tf.RunMetadata()
#_,G_current,output = sess.run([G_opt,G_loss,out],feed_dict={in_tile:input_patch,gt:gt_patch,lr:LR},options=run_options,run_metadata=run_metadata)
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
_,G_current,output = sess.run([G_opt,G_loss,out],feed_dict={in_tile:input_patch,gt:gt_patch,lr:LR})
_,G_current,output,summary = sess.run([G_opt,G_loss,out,merged],feed_dict={in_tile:input_patch,gt:gt_patch,lr:LR},options=run_options,run_metadata=run_metadata)
#_,G_current,output = sess.run([G_opt,G_loss,out],feed_dict={in_tile:input_patch,gt:gt_patch,lr:LR})
g_loss[i]=G_current
mean_loss = np.mean(g_loss[np.where(g_loss)])
g_loss[i]=G_current
mean_loss = np.mean(g_loss[np.where(g_loss)])
if DEBUG_PLT_LOSS:
if DEBUG_PLT_LOSS:
recorded_loss.append(G_current)
recorded_mean_loss.append(mean_loss)
recorded_loss.append(G_current)
recorded_mean_loss.append(mean_loss)
recorded_pr_d.append(output[0,0])
recorded_pr_c.append(output[0,1])
recorded_pr_d.append(output[0,0])
recorded_pr_c.append(output[0,1])
recorded_gt_d.append(gt_patch[0,0])
recorded_gt_c.append(gt_patch[0,1])
recorded_gt_d.append(gt_patch[0,0])
recorded_gt_c.append(gt_patch[0,1])
plt.clf()
plt.clf()
plt.subplot(311)
plt.subplot(311)
plt.plot(recorded_loss, label='loss')
plt.plot(recorded_mean_loss, label='mean loss', color='red')
plt.xlabel('Iteration')
plt.ylabel('Loss')
plt.title("Loss=%.5f, Mean Loss=%.5f"%(G_current,mean_loss), fontdict={'size': 20, 'color': 'red'})
#plt.text(0.5, 0.5, 'Loss=%.5f' % G_current, fontdict={'size': 20, 'color': 'red'})
plt.plot(recorded_loss, label='loss')
plt.plot(recorded_mean_loss, label='mean loss', color='red')
plt.xlabel('Iteration')
plt.ylabel('Loss')
plt.title("Loss=%.5f, Mean Loss=%.5f"%(G_current,mean_loss), fontdict={'size': 20, 'color': 'red'})
#plt.text(0.5, 0.5, 'Loss=%.5f' % G_current, fontdict={'size': 20, 'color': 'red'})
plt.subplot(312)
plt.subplot(312)
plt.xlabel('Iteration')
plt.ylabel('Disparities')
plt.plot(recorded_gt_d, label='gt_d',color='green')
plt.plot(recorded_pr_d, label='pr_d',color='red')
plt.legend(loc='best',ncol=1)
plt.xlabel('Iteration')
plt.ylabel('Disparities')
plt.plot(recorded_gt_d, label='gt_d',color='green')
plt.plot(recorded_pr_d, label='pr_d',color='red')
plt.legend(loc='best',ncol=1)
plt.subplot(313)
plt.subplot(313)
plt.xlabel('Iteration')
plt.ylabel('Confidences')
plt.plot(recorded_gt_c, label='gt_c',color='green')
plt.plot(recorded_pr_c, label='pr_c',color='red')
plt.legend(loc='best',ncol=1)
plt.xlabel('Iteration')
plt.ylabel('Confidences')
plt.plot(recorded_gt_c, label='gt_c',color='green')
plt.plot(recorded_pr_c, label='pr_c',color='red')
plt.legend(loc='best',ncol=1)
plt.pause(0.001)
plt.pause(0.001)
else:
print("%d %d Loss=%.3f CurrentLoss=%.3f Time=%.3f"%(epoch,i,mean_loss,G_current,time.time()-st))
else:
print("%d %d Loss=%.3f CurrentLoss=%.3f Time=%.3f"%(epoch,i,mean_loss,G_current,time.time()-st))
#train_writer.add_run_metadata(run_metadata, 'step%d' % cnt)
#test_writer.add_summary(summary,cnt)
#train_writer.add_summary(summary, cnt)
if epoch%save_freq==0:
if not os.path.isdir(result_dir + '%04d'%epoch):
os.makedirs(result_dir + '%04d'%epoch)
if epoch%save_freq==0:
if not os.path.isdir(result_dir + '%04d'%epoch):
os.makedirs(result_dir + '%04d'%epoch)
saver.save(sess, checkpoint_dir + 'model.ckpt')
train_writer.close()
test_writer.close()
print_time()
print(bcolors.OKGREEN+"time: "+str(time.time())+bcolors.ENDC)
......
test_nn_infer.py
View file @ 0452a446
......@@ -44,6 +44,7 @@ def print_time():
VALUES_LAYER_NAME = 'other'
LAYERS_OF_INTEREST = ['diagm-pair', 'diago-pair', 'hor-pairs', 'vert-pairs']
RADIUS = 1
TILE_PACKING_TYPE = 1
try:
src = sys.argv[1]
......@@ -83,7 +84,11 @@ def network(input):
sess = tf.Session()
in_tile = tf.placeholder(tf.float32,[None,101])
if TILE_PACKING_TYPE==1:
in_tile = tf.placeholder(tf.float32,[None,101])
elif TILE_PACKING_TYPE==2:
in_tile = tf.placeholder(tf.float32,[None,105])
gt = tf.placeholder(tf.float32,[None,2])
out = network(in_tile)
......@@ -143,7 +148,11 @@ for item in tlist:
# tiles and values
# might not need it because going to loop through anyway
packed_tiles = pile.pack(tiles)
if TILE_PACKING_TYPE==1:
packed_tiles = pile.pack(tiles)
elif TILE_PACKING_TYPE==2:
packed_tiles = pile.pack(tiles,TILE_PACKING_TYPE)
packed_tiles = np.dstack((packed_tiles,values[:,:,0]))
print(packed_tiles.shape)
......@@ -168,11 +177,12 @@ for item in tlist:
packed_tiles_flat = packed_tiles[i]
values_flat = values[i]
# whole row at once
output = sess.run(out,feed_dict={in_tile:packed_tiles_flat})
output_image[i] = output
# so, let's print
for j in range(output.shape[0]):
for j in range(packed_tiles.shape[0]):
p = output[j,0]
pc = output[j,1]
fv = values_flat[j,0]
......@@ -204,6 +214,15 @@ for item in tlist:
tif = np.dstack((im1,im2,im3))
im3 = np.ravel(im3)
print(im3.shape)
im4 = im3[~np.isnan(im3)]
rms = np.sqrt(np.mean(np.square(im4)))
print("RMS = "+str(rms))
imagej_tiffwriter.save('prediction_results.tiff',tif)
#sys.exit(0)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment