I am using the DeepLabV3+ repository and I noticed that the loss_weight is set to 1.0 which means that we are weighting classes in the same way. However, I have a really unbalanced dataset, something like 80% negative and 20% positive.
def add_softmax_cross_entropy_loss_for_each_scale(scales_to_logits,
labels,
num_classes,
ignore_label,
loss_weight=1.0,
upsample_logits=True,
hard_example_mining_step=0,
top_k_percent_pixels=1.0,
scope=None):
"""Adds softmax cross entropy loss for logits of each scale.
Args:
scales_to_logits: A map from logits names for different scales to logits.
The logits have shape [batch, logits_height, logits_width, num_classes].
labels: Groundtruth labels with shape [batch, image_height, image_width, 1].
num_classes: Integer, number of target classes.
ignore_label: Integer, label to ignore.
loss_weight: Float, loss weight.
upsample_logits: Boolean, upsample logits or not.
hard_example_mining_step: An integer, the training step in which the hard
exampling mining kicks off. Note that we gradually reduce the mining
percent to the top_k_percent_pixels. For example, if
hard_example_mining_step = 100K and top_k_percent_pixels = 0.25, then
mining percent will gradually reduce from 100% to 25% until 100K steps
after which we only mine top 25% pixels.
top_k_percent_pixels: A float, the value lies in [0.0, 1.0]. When its value
< 1.0, only compute the loss for the top k percent pixels (e.g., the top
20% pixels). This is useful for hard pixel mining.
scope: String, the scope for the loss.
Raises:
ValueError: Label or logits is None.
"""
if labels is None:
raise ValueError('No label for softmax cross entropy loss.')
for scale, logits in six.iteritems(scales_to_logits):
loss_scope = None
if scope:
loss_scope = '%s_%s' % (scope, scale)
if upsample_logits:
# Label is not downsampled, and instead we upsample logits.
logits = tf.image.resize_bilinear(
logits,
preprocess_utils.resolve_shape(labels, 4)[1:3],
align_corners=True)
scaled_labels = labels
else:
# Label is downsampled to the same size as logits.
scaled_labels = tf.image.resize_nearest_neighbor(
labels,
preprocess_utils.resolve_shape(logits, 4)[1:3],
align_corners=True)
scaled_labels = tf.reshape(scaled_labels, shape=[-1])
not_ignore_mask = tf.to_float(tf.not_equal(scaled_labels,
ignore_label)) * loss_weight
one_hot_labels = tf.one_hot(
scaled_labels, num_classes, on_value=1.0, off_value=0.0)
if top_k_percent_pixels == 1.0:
# Compute the loss for all pixels.
tf.compat.v1.losses.softmax_cross_entropy(
one_hot_labels,
tf.reshape(logits, shape=[-1, num_classes]),
weights=not_ignore_mask,
scope=loss_scope)
else:
logits = tf.reshape(logits, shape=[-1, num_classes])
weights = not_ignore_mask
with tf.name_scope(loss_scope, 'softmax_hard_example_mining',
[logits, one_hot_labels, weights]):
one_hot_labels = tf.stop_gradient(
one_hot_labels, name='labels_stop_gradient')
pixel_losses = tf.nn.softmax_cross_entropy_with_logits_v2(
labels=one_hot_labels,
logits=logits,
name='pixel_losses')
weighted_pixel_losses = tf.multiply(pixel_losses, weights)
num_pixels = tf.to_float(tf.shape(logits)[0])
# Compute the top_k_percent pixels based on current training step.
if hard_example_mining_step == 0:
# Directly focus on the top_k pixels.
top_k_pixels = tf.to_int32(top_k_percent_pixels * num_pixels)
else:
# Gradually reduce the mining percent to top_k_percent_pixels.
global_step = tf.to_float(tf.train.get_or_create_global_step())
ratio = tf.minimum(1.0, global_step / hard_example_mining_step)
top_k_pixels = tf.to_int32(
(ratio * top_k_percent_pixels + (1.0 - ratio)) * num_pixels)
top_k_losses, _ = tf.nn.top_k(weighted_pixel_losses,
k=top_k_pixels,
sorted=True,
name='top_k_percent_pixels')
total_loss = tf.reduce_sum(top_k_losses)
num_present = tf.reduce_sum(
tf.to_float(tf.not_equal(top_k_losses, 0.0)))
loss = _div_maybe_zero(total_loss, num_present)
tf.losses.add_loss(loss)
This is the loss function they use and as you can see the loss_weight is 1.0.
not_ignore_mask = tf.to_float(tf.not_equal(scaled_labels,
ignore_label)) * loss_weight
I want to give a weight of 0.2 to negative classes and 0.8 to my positve class (predictions). Does anyone know how to do this or any repo/example that have done this before?
regards
you can change the weights in "train_utils.py" inside "utils" folder.
In "def add_softmax_cross_entropy_loss_for_each_scale(...)"
Something like this:
for scale, logits in six.iteritems(scales_to_logits):
loss_scope = None
irgore_weight = 0
label0_weight = 0.2 #I don't know your labels order...
label1_weight = 0.8 #I don't know your labels order...
Also change the not_ignore_mask, like this:
not_ignore_mask = tf.to_float(tf.equal(scaled_labels, 0)) * label0_weight + tf.to_float(tf.equal(scaled_labels, 1)) * label1_weight + tf.to_float(tf.equal(scaled_labels, ignore_label)) * irgore_weight
I hope it is helpful.