fix sampler bugs and update dataloader

Author: bubbliiiing

train.py

@@ -427,8 +427,11 @@
 
                 UnFreeze_flag = True
 
+            if distributed:
+                train_sampler.set_epoch(epoch)
             set_optimizer_lr(optimizer, lr_scheduler_func, epoch)
 
             fit_one_epoch(model_train, model, yolo_loss, loss_history, optimizer, epoch, epoch_step, epoch_step_val, gen, gen_val, UnFreeze_Epoch, Cuda, fp16, scaler, save_period, save_dir, local_rank)
-
-        loss_history.writer.close()
+            
+        if local_rank == 0:
+            loss_history.writer.close()

utils/dataloader.py

@@ -1,5 +1,6 @@
 import cv2
 import numpy as np
+import torch
 from PIL import Image
 from torch.utils.data.dataset import Dataset
 
@@ -160,7 +161,6 @@ def yolo_dataset_collate(batch):
     for img, box in batch:
         images.append(img)
         bboxes.append(box)
-    images = np.array(images)
+    images = torch.from_numpy(np.array(images)).type(torch.FloatTensor)
+    bboxes = [torch.from_numpy(ann).type(torch.FloatTensor) for ann in bboxes]
     return images, bboxes
-
-

utils/utils_fit.py

@@ -21,11 +21,8 @@ def fit_one_epoch(model_train, model, yolo_loss, loss_history, optimizer, epoch,
         images, targets = batch[0], batch[1]
         with torch.no_grad():
             if cuda:
-                images  = torch.from_numpy(images).type(torch.FloatTensor).cuda()
-                targets = [torch.from_numpy(ann).type(torch.FloatTensor).cuda() for ann in targets]
-            else:
-                images  = torch.from_numpy(images).type(torch.FloatTensor)
-                targets = [torch.from_numpy(ann).type(torch.FloatTensor) for ann in targets]
+                images  = images.cuda()
+                targets = [ann.cuda() for ann in targets]
         #----------------------#
         #   清零梯度
         #----------------------#
@@ -94,11 +91,8 @@ def fit_one_epoch(model_train, model, yolo_loss, loss_history, optimizer, epoch,
         images, targets = batch[0], batch[1]
         with torch.no_grad():
             if cuda:
-                images  = torch.from_numpy(images).type(torch.FloatTensor).cuda()
-                targets = [torch.from_numpy(ann).type(torch.FloatTensor).cuda() for ann in targets]
-            else:
-                images  = torch.from_numpy(images).type(torch.FloatTensor)
-                targets = [torch.from_numpy(ann).type(torch.FloatTensor) for ann in targets]
+                images  = images.cuda()
+                targets = [ann.cuda() for ann in targets]
             #----------------------#
             #   清零梯度
             #----------------------#