Fix ClearML Mosaic callback to 'on_train_epoch_end' (#92)

ultralytics/yolo/engine/validator.py

@@ -47,6 +47,7 @@ class BaseValidator:
             model = model.half() if self.args.half else model.float()
             self.model = model
             self.loss = torch.zeros_like(trainer.loss_items, device=trainer.device)
+            self.args.plots = trainer.epoch == trainer.epochs - 1  # always plot final epoch
         else:
             assert model is not None, "Either trainer or model is needed for validation"
             self.device = select_device(self.args.device, self.args.batch_size)

ultralytics/yolo/utils/callbacks/clearml.py

@@ -1,6 +1,3 @@
-import os
-from pathlib import Path
-
 from ultralytics.yolo.utils.torch_utils import get_flops, get_num_params
 
 try:
@@ -30,12 +27,9 @@ def on_train_start(trainer):
     task.connect(dict(trainer.args), name='General')
 
 
-def on_epoch_start(trainer):
+def on_train_epoch_end(trainer):
     if trainer.epoch == 1:
-        plots = [filename for filename in os.listdir(trainer.save_dir) if filename.startswith("train_batch")]
+        _log_images({f.stem: str(f) for f in trainer.save_dir.glob('train_batch*.jpg')}, "Mosaic", trainer.epoch)
-        imgs_dict = {f"train_batch_{i}": Path(trainer.save_dir) / img for i, img in enumerate(plots)}
-        if imgs_dict:
-            _log_images(imgs_dict, "Mosaic", trainer.epoch)
 
 
 def on_val_end(trainer):
@@ -55,6 +49,6 @@ def on_train_end(trainer):
 
 callbacks = {
     "on_train_start": on_train_start,
-    "on_epoch_start": on_epoch_start,
+    "on_train_epoch_end": on_train_epoch_end,
     "on_val_end": on_val_end,
     "on_train_end": on_train_end} if clearml else {}

ultralytics/yolo/utils/metrics.py

@@ -343,7 +343,7 @@ def compute_ap(recall, precision):
     return ap, mpre, mrec
 
 
-def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16, prefix=""):
+def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir=Path(), names=(), eps=1e-16, prefix=""):
     """ Compute the average precision, given the recall and precision curves.
     Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
     # Arguments
@@ -398,10 +398,10 @@ def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names
     names = [v for k, v in names.items() if k in unique_classes]  # list: only classes that have data
     names = dict(enumerate(names))  # to dict
     if plot:
-        plot_pr_curve(px, py, ap, Path(save_dir) / f'{prefix}PR_curve.png', names)
+        plot_pr_curve(px, py, ap, save_dir / f'{prefix}PR_curve.png', names)
-        plot_mc_curve(px, f1, Path(save_dir) / f'{prefix}F1_curve.png', names, ylabel='F1')
+        plot_mc_curve(px, f1, save_dir / f'{prefix}F1_curve.png', names, ylabel='F1')
-        plot_mc_curve(px, p, Path(save_dir) / f'{prefix}P_curve.png', names, ylabel='Precision')
+        plot_mc_curve(px, p, save_dir / f'{prefix}P_curve.png', names, ylabel='Precision')
-        plot_mc_curve(px, r, Path(save_dir) / f'{prefix}R_curve.png', names, ylabel='Recall')
+        plot_mc_curve(px, r, save_dir / f'{prefix}R_curve.png', names, ylabel='Recall')
 
     i = smooth(f1.mean(0), 0.1).argmax()  # max F1 index
     p, r, f1 = p[:, i], r[:, i], f1[:, i]

ultralytics/yolo/v8/detect/train.py

@@ -63,16 +63,15 @@ class DetectionTrainer(BaseTrainer):
         return dict(zip(keys, loss_items)) if loss_items is not None else keys
 
     def progress_string(self):
-        return ('\n' + '%11s' * 7) % \
+        return ('\n' + '%11s' * 7) % ('Epoch', 'GPU_mem', *self.loss_names, 'Instances', 'Size')
-               ('Epoch', 'GPU_mem', *self.loss_names, 'Instances', 'Size')
 
     def plot_training_samples(self, batch, ni):
-        images = batch["img"]
+        plot_images(images=batch["img"],
-        cls = batch["cls"].squeeze(-1)
+                    batch_idx=batch["batch_idx"],
-        bboxes = batch["bboxes"]
+                    cls=batch["cls"].squeeze(-1),
-        paths = batch["im_file"]
+                    bboxes=batch["bboxes"],
-        batch_idx = batch["batch_idx"]
+                    paths=batch["im_file"],
-        plot_images(images, batch_idx, cls, bboxes, paths=paths, fname=self.save_dir / f"train_batch{ni}.jpg")
+                    fname=self.save_dir / f"train_batch{ni}.jpg")
 
     def plot_metrics(self):
         plot_results(file=self.csv)  # save results.png

ultralytics/yolo/v8/segment/train.py

@@ -214,8 +214,7 @@ class SegmentationTrainer(DetectionTrainer):
         return dict(zip(keys, loss_items)) if loss_items is not None else keys
 
     def progress_string(self):
-        return ('\n' + '%11s' * 8) % \
+        return ('\n' + '%11s' * 8) % ('Epoch', 'GPU_mem', *self.loss_names, 'Instances', 'Size')
-               ('Epoch', 'GPU_mem', *self.loss_names, 'Instances', 'Size')
 
     def plot_training_samples(self, batch, ni):
         images = batch["img"]