From 0f413ac4a0f9585636552d26a82e05d2de4b518f Mon Sep 17 00:00:00 2001
From: Andre Anjos <andre.dos.anjos@gmail.com>
Date: Wed, 13 May 2020 15:55:39 +0200
Subject: [PATCH] [engine.ssltrainer] Validation does not compute SSL loss,
just standard model performance
---
bob/ip/binseg/engine/ssltrainer.py | 29 ++++++++++-------------------
1 file changed, 10 insertions(+), 19 deletions(-)
diff --git a/bob/ip/binseg/engine/ssltrainer.py b/bob/ip/binseg/engine/ssltrainer.py
index 6d378268..4a8cca16 100644
--- a/bob/ip/binseg/engine/ssltrainer.py
+++ b/bob/ip/binseg/engine/ssltrainer.py
@@ -376,6 +376,7 @@ def run(
scheduler.step()
# calculates the validation loss if necessary
+ # note: validation does not comprise "unlabelled" losses
valid_losses = None
if valid_loader is not None:
@@ -385,8 +386,7 @@ def run(
for samples in tqdm(
valid_loader, desc="valid", leave=False, disable=None
):
-
- # labelled
+ # data forwarding on the existing network
images = samples[1].to(
device=device,
non_blocking=torch.cuda.is_available(),
@@ -395,25 +395,16 @@ def run(
device=device,
non_blocking=torch.cuda.is_available(),
)
- unlabelled_images = samples[4].to(
- device=device,
- non_blocking=torch.cuda.is_available(),
- )
- # labelled outputs
+ masks = None
+ if len(samples) == 4:
+ masks = samples[-1].to(
+ device=device,
+ non_blocking=torch.cuda.is_available(),
+ )
+
outputs = model(images)
- unlabelled_outputs = model(unlabelled_images)
- # guessed unlabelled outputs
- unlabelled_ground_truths = guess_labels(
- unlabelled_images, model
- )
- loss, ll, ul = criterion(
- outputs,
- ground_truths,
- unlabelled_outputs,
- unlabelled_ground_truths,
- ramp_up_factor,
- )
+ loss = criterion(outputs, ground_truths, masks)
valid_losses.update(loss)
if checkpoint_period and (epoch % checkpoint_period == 0):
--
GitLab