Bienqq
/
car-detection-bayes
2
1
Fork 0
Code Issues Pull Requests Releases Wiki Activity

tensorboard updates

This commit is contained in:
Glenn Jocher 2019-08-09 16:37:19 +02:00
parent fdd5afa229
commit 933f85f632
3 changed files with 22 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
requirements.txt
View File

@ -1,6 +1,8 @@
# pip3 install -U -r requirements.txt # pip3 install -U -r requirements.txt
# conda install -y numpy opencv matplotlib tqdm pillow && conda install -y scikit-image -c conda-forge && conda install -y -c spyder-ide spyder-line-profiler # conda install -y numpy opencv matplotlib tqdm pillow && conda install -y scikit-image -c conda-forge && conda install -y -c spyder-ide spyder-line-profiler
# conda install -y -c conda-forge tensorboard && conda install -y -c anaconda future
# conda install pytorch torchvision -c pytorch # conda install pytorch torchvision -c pytorch
numpy numpy
opencv-python opencv-python
torch >= 1.1.0 torch >= 1.1.0

49
train.py
View File

@ -59,8 +59,7 @@ def train(cfg,
img_size=416, img_size=416,
epochs=100, # 500200 batches at bs 16, 117263 images = 273 epochs epochs=100, # 500200 batches at bs 16, 117263 images = 273 epochs
batch_size=16, batch_size=16,
accumulate=4, accumulate=4): # effective bs = batch_size * accumulate = 16 * 4 = 64
write_to_tensorboard=False): # effective bs = batch_size * accumulate = 16 * 4 = 64
# Initialize # Initialize
init_seeds() init_seeds()
weights = 'weights' + os.sep weights = 'weights' + os.sep
@ -227,10 +226,10 @@ def train(cfg,
# Plot images with bounding boxes # Plot images with bounding boxes
if epoch == 0 and i == 0: if epoch == 0 and i == 0:
figure_image = plot_images(imgs=imgs, targets=targets, paths=paths, fname='train_batch%g.jpg' % i) fname = 'train_batch%g.jpg' % i
fig_data = plot_images(imgs=imgs, targets=targets, paths=paths, fname=fname)
if write_to_tensorboard: if tb_writer:
tb_writer.add_image('train_batch', figure_image, dataformats='HWC') tb_writer.add_image(fname, fig_data, dataformats='HWC')
# Hyperparameter burn-in # Hyperparameter burn-in
# n_burn = nb - 1 # min(nb // 5 + 1, 1000) # number of burn-in batches # n_burn = nb - 1 # min(nb // 5 + 1, 1000) # number of burn-in batches
@ -282,19 +281,12 @@ def train(cfg,
file.write(s + '%11.3g' * 7 % results + '\n') # P, R, mAP, F1, test_losses=(GIoU, obj, cls) file.write(s + '%11.3g' * 7 % results + '\n') # P, R, mAP, F1, test_losses=(GIoU, obj, cls)
# Write Tensorboard results # Write Tensorboard results
if write_to_tensorboard: if tb_writer:
tb_writer.add_scalar('GIoU/XY', mloss[0], epoch) x = list(mloss[:5]) + list(results[:7])
tb_writer.add_scalar('Width/Height', mloss[1], epoch) titles = ['GIoU/XY', 'Width/Height', 'Objectness', 'Classification', 'Train loss', 'Precision', 'Recall',
tb_writer.add_scalar('Confidence', mloss[2], epoch) 'mAP', 'F1', 'val GIoU', 'val Objectness', 'val Classification']
tb_writer.add_scalar('Classification', mloss[3], epoch) for xi, title in zip(x, titles):
tb_writer.add_scalar('Train loss', mloss[4], epoch) tb_writer.add_scalar(title, xi, epoch)
tb_writer.add_scalar('Precision', results[0], epoch)
tb_writer.add_scalar('Recall', results[1], epoch)
tb_writer.add_scalar('mAP', results[2], epoch)
tb_writer.add_scalar('F1', results[3], epoch)
tb_writer.add_scalar('Test loss GIoU', results[4], epoch)
tb_writer.add_scalar('Test loss obj', results[5], epoch)
tb_writer.add_scalar('Test loss cls', results[6], epoch)
# Update best map # Update best map
fitness = results[2] # mAP fitness = results[2] # mAP
@ -360,25 +352,20 @@ if __name__ == '__main__':
print(opt) print(opt)
if not opt.evolve: # Train normally if not opt.evolve: # Train normally
# Tensorboard support, try:
# start with "tensorboard --logdir=runs" then go to localhost:6006 # Start Tensorboard with "tensorboard --logdir=runs", view at http://localhost:6006/
tensorboard_support = True from torch.utils.tensorboard import SummaryWriter
if version_to_tuple(torch.__version__) >= version_to_tuple("1.1.0"):
try:
from torch.utils.tensorboard import SummaryWriter
tb_train_name = time.time() tb_writer = SummaryWriter()
tb_writer = SummaryWriter('runs/{}'.format(tb_train_name)) except:
except: tb_writer = None
tensorboard_support = False
results = train(opt.cfg, results = train(opt.cfg,
opt.data, opt.data,
img_size=opt.img_size, img_size=opt.img_size,
epochs=opt.epochs, epochs=opt.epochs,
batch_size=opt.batch_size, batch_size=opt.batch_size,
accumulate=opt.accumulate, accumulate=opt.accumulate)
write_to_tensorboard=tensorboard_support)
else: # Evolve hyperparameters (optional) else: # Evolve hyperparameters (optional)
opt.notest = True # only test final epoch opt.notest = True # only test final epoch

16
utils/utils.py
View File

@ -819,18 +819,6 @@ def version_to_tuple(version):
def fig_to_data(fig): def fig_to_data(fig):
# Used to convert a Matplotlib figure to a 4D numpy array with RGBA channels and return it # Converts a matplotlib fig to 3D numpy array (fig is a matplotlib figure)
# param fig a matplotlib figure
# return a numpy 3D array of RGBA values
# draw the renderer
fig.canvas.draw() fig.canvas.draw()
return np.array(fig.canvas.renderer.buffer_rgba())[:, :, :3] # RGB image
# Get the RGBA buffer from the figure
w,h = fig.canvas.get_width_height()
buf = np.fromstring (fig.canvas.tostring_argb(), dtype=np.uint8)
buf.shape = (w, h, 4)
# canvas.tostring_argb give pixmap in ARGB mode. Roll the ALPHA channel to have it in RGBA mode
buf = np.roll(buf, 3, axis = 2)
return buf[:,:,:3] # Return RGB numpy image