Merge pull request WZMIAOMIAO#186 from WZMIAOMIAO/dev

Dev

deploying_service/deploying_pytorch/pytorch_flask_service/requirements.txt

@@ -1,3 +1,3 @@
 Flask==1.1.1
 Flask_Cors==3.0.9
-Pillow==7.2.0
+Pillow==8.1.1

pytorch_classification/tensorboard_test/requirements.txt

@@ -2,5 +2,5 @@ torchvision==0.7.0
 tqdm==4.42.1
 matplotlib==3.2.1
 torch==1.6.0
-Pillow==8.0.1
+Pillow==8.1.1
 tensorboard==2.2.2

pytorch_object_detection/faster_rcnn/requirements.txt

@@ -5,4 +5,4 @@ tqdm==4.42.1
 torch==1.6.0
 torchvision==0.7.0
 pycocotools==2.0.0
-Pillow==8.0.1
+Pillow==8.1.1

pytorch_object_detection/ssd/plot_curve.py

@@ -6,7 +6,7 @@ def plot_loss_and_lr(train_loss, learning_rate):
         x = list(range(len(train_loss)))
         fig, ax1 = plt.subplots(1, 1)
         ax1.plot(x, train_loss, 'r', label='loss')
-        ax1.set_xlabel("step")
+        ax1.set_xlabel("epoch")
         ax1.set_ylabel("loss")
         ax1.set_title("Train Loss and lr")
         plt.legend(loc='best')

pytorch_object_detection/ssd/predict_test.py

@@ -37,9 +37,6 @@ def main():
 
     model.load_state_dict(train_weights_dict, strict=False)
     model.to(device)
-    # initial model
-    init_img = torch.zeros((1, 3, 300, 300), device=device)
-    model(init_img)
 
     # read class_indict
     json_path = "./pascal_voc_classes.json"
@@ -61,6 +58,10 @@ def main():
 
     model.eval()
     with torch.no_grad():
+        # initial model
+        init_img = torch.zeros((1, 3, 300, 300), device=device)
+        model(init_img)
+
         time_start = time_synchronized()
         predictions = model(img.to(device))[0]  # bboxes_out, labels_out, scores_out
         time_end = time_synchronized()

pytorch_object_detection/ssd/requirements.txt

@@ -5,4 +5,4 @@ pycocotools==2.0.0
 torch==1.6.0
 torchvision==0.7.0
 lxml==4.6.2
-Pillow==8.0.1
+Pillow==8.1.1

pytorch_object_detection/ssd/src/utils.py

@@ -590,13 +590,15 @@ def decode_single_new(self, bboxes_in, scores_in, criteria, num_output):
 
         # remove low scoring boxes
         # 移除低概率目标，self.scores_thresh=0.05
-        inds = torch.nonzero(scores_in > 0.05).squeeze(1)
+        # inds = torch.nonzero(scores_in > 0.05).squeeze(1)
+        inds = torch.where(torch.gt(scores_in, 0.05))[0]
         bboxes_in, scores_in, labels = bboxes_in[inds, :], scores_in[inds], labels[inds]
 
         # remove empty boxes
         ws, hs = bboxes_in[:, 2] - bboxes_in[:, 0], bboxes_in[:, 3] - bboxes_in[:, 1]
         keep = (ws >= 1 / 300) & (hs >= 1 / 300)
-        keep = keep.nonzero().squeeze(1)
+        # keep = keep.nonzero().squeeze(1)
+        keep = torch.where(keep)[0]
         bboxes_in, scores_in, labels = bboxes_in[keep], scores_in[keep], labels[keep]
 
         # non-maximum suppression

pytorch_object_detection/ssd/train_ssd300.py

@@ -1,4 +1,5 @@
 import os
+import datetime
 
 import torch
 
@@ -45,6 +46,8 @@ def main(parser_data):
     if not os.path.exists("save_weights"):
         os.mkdir("save_weights")
 
+    results_file = "results{}.txt".format(datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
+
     data_transform = {
         "train": transform.Compose([transform.SSDCropping(),
                                     transform.Resize(),
@@ -113,16 +116,26 @@ def main(parser_data):
     # 提前加载验证集数据，以免每次验证时都要重新加载一次数据，节省时间
     val_data = get_coco_api_from_dataset(val_data_loader.dataset)
     for epoch in range(parser_data.start_epoch, parser_data.epochs):
-        utils.train_one_epoch(model=model, optimizer=optimizer,
-                              data_loader=train_data_loader,
-                              device=device, epoch=epoch,
-                              print_freq=50, train_loss=train_loss,
-                              train_lr=learning_rate)
-
+        mean_loss, lr = utils.train_one_epoch(model=model, optimizer=optimizer,
+                                              data_loader=train_data_loader,
+                                              device=device, epoch=epoch,
+                                              print_freq=50)
+        train_loss.append(mean_loss.item())
+        learning_rate.append(lr)
+
+        # update learning rate
         lr_scheduler.step()
 
-        utils.evaluate(model=model, data_loader=val_data_loader,
-                       device=device, data_set=val_data, mAP_list=val_map)
+        coco_info = utils.evaluate(model=model, data_loader=val_data_loader,
+                                   device=device, data_set=val_data)
+
+        # write into txt
+        with open(results_file, "a") as f:
+            result_info = [str(round(i, 4)) for i in coco_info + [mean_loss.item(), lr]]
+            txt = "epoch:{} {}".format(epoch, '  '.join(result_info))
+            f.write(txt + "\n")
+
+        val_map.append(coco_info[1])  # pascal mAP
 
         # save weights
         save_files = {

pytorch_object_detection/ssd/train_utils/train_eval_utils.py

@@ -10,8 +10,8 @@
 import train_utils.distributed_utils as utils
 
 
-def train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq,
-                    train_loss=None, train_lr=None, warmup=False):
+def train_one_epoch(model, optimizer, data_loader, device, epoch,
+                    print_freq=50, warmup=False):
     model.train()
     metric_logger = utils.MetricLogger(delimiter="  ")
     metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
@@ -24,7 +24,8 @@ def train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq,
 
         lr_scheduler = utils.warmup_lr_scheduler(optimizer, warmup_iters, warmup_factor)
 
-    for images, targets in metric_logger.log_every(data_loader, print_freq, header):
+    mloss = torch.zeros(1).to(device)  # mean losses
+    for i, [images, targets] in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
         # batch inputs information
         images = torch.stack(images, dim=0)
 
@@ -49,10 +50,9 @@ def train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq,
         losses_dict_reduced = utils.reduce_dict(losses_dict)
         losses_reduce = losses_dict_reduced["total_losses"]
 
-        loss_value = losses_reduce.item()
-        if isinstance(train_loss, list):
-            # 记录训练损失
-            train_loss.append(loss_value)
+        loss_value = losses_reduce.detach()
+        # 记录训练损失
+        mloss = (mloss * i + loss_value) / (i + 1)  # update mean losses
 
         if not math.isfinite(loss_value):  # 当计算的损失为无穷大时停止训练
             print("Loss is {}, stopping training".format(loss_value))
@@ -70,12 +70,12 @@ def train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq,
         metric_logger.update(**losses_dict_reduced)
         now_lr = optimizer.param_groups[0]["lr"]
         metric_logger.update(lr=now_lr)
-        if isinstance(train_lr, list):
-            train_lr.append(now_lr)
+
+    return mloss, now_lr
 
 
 @torch.no_grad()
-def evaluate(model, data_loader, device, data_set=None, mAP_list=None):
+def evaluate(model, data_loader, device, data_set=None):
     n_threads = torch.get_num_threads()
     # FIXME remove this and make paste_masks_in_image run on the GPU
     torch.set_num_threads(1)
@@ -138,12 +138,9 @@ def evaluate(model, data_loader, device, data_set=None, mAP_list=None):
     coco_evaluator.summarize()
     torch.set_num_threads(n_threads)
 
-    print_txt = coco_evaluator.coco_eval[iou_types[0]].stats
-    coco_mAP = print_txt[0]
-    voc_mAP = print_txt[1]
-    if isinstance(mAP_list, list):
-        mAP_list.append(voc_mAP)
-    # return coco_evaluator
+    coco_info = coco_evaluator.coco_eval[iou_types[0]].stats.tolist()  # numpy to list
+
+    return coco_info
 
 
 def _get_iou_types(model):