From 8501aed49fc5d617db685b8d80fcd5460d768841 Mon Sep 17 00:00:00 2001
From: Glenn Jocher <glenn.jocher@ultralytics.com>
Date: Mon, 15 Jul 2019 17:54:31 +0200
Subject: [PATCH] updates

---
 README.md         |  7 ++++---
 train.py          | 14 +++++++-------
 utils/datasets.py | 10 +++++++---
 utils/gcp.sh      | 12 ++++++------
 4 files changed, 24 insertions(+), 19 deletions(-)

diff --git a/README.md b/README.md
index 001668ee..8d619d8e 100755
--- a/README.md
+++ b/README.md
@@ -50,7 +50,7 @@ https://colab.research.google.com/drive/1G8T-VFxQkjDe4idzN8F-hbIBqkkkQnxw
 
 **Start Training:** `python3 train.py` to begin training after downloading COCO data with `data/get_coco_dataset.sh`.
 
-**Resume Training:** `python3 train.py --resume` to resume training from `weights/latest.pt`.
+**Resume Training:** `python3 train.py --resume` to resume training from `weights/last.pt`.
 
 Each epoch trains on 117,263 images from the train and validate COCO sets, and tests on 5000 images from the COCO validate set. Default training settings produce loss plots below, with **training speed of 0.25 s/batch on a V100 GPU (almost 50 COCO epochs/day)**.
 
@@ -136,8 +136,9 @@ Success: converted 'weights/yolov3-spp.pt' to 'converted.weights'
 
 # mAP
 
-- Use `test.py --weights weights/yolov3.weights` to test the official YOLOv3 weights.
-- Use `test.py --weights weights/latest.pt` to test the latest training results.
+- `test.py --weights weights/yolov3.weights` to test official YOLOv3 weights.
+- `test.py --weights weights/last.pt` to test most recent checkpoint.
+- `test.py --weights weights/best.pt` to test best checkpoint.
 - Compare to darknet published results https://arxiv.org/abs/1804.02767.
 
 <!---
diff --git a/train.py b/train.py
index 2504265b..68e0c3ca 100644
--- a/train.py
+++ b/train.py
@@ -45,7 +45,7 @@ def train(cfg,
     # Initialize
     init_seeds()
     weights = 'weights' + os.sep
-    latest = weights + 'latest.pt'
+    last = weights + 'last.pt'
     best = weights + 'best.pt'
     device = torch_utils.select_device()
     multi_scale = opt.multi_scale
@@ -79,10 +79,10 @@ def train(cfg,
             for p in model.parameters():
                 p.requires_grad = True if p.shape[0] == nf else False
 
-        else:  # resume from latest.pt
+        else:  # resume from last.pt
             if opt.bucket:
-                os.system('gsutil cp gs://%s/latest.pt %s' % (opt.bucket, latest))  # download from bucket
-            chkpt = torch.load(latest, map_location=device)  # load checkpoint
+                os.system('gsutil cp gs://%s/last.pt %s' % (opt.bucket, last))  # download from bucket
+            chkpt = torch.load(last, map_location=device)  # load checkpoint
             model.load_state_dict(chkpt['model'])
 
         if chkpt['optimizer'] is not None:
@@ -273,10 +273,10 @@ def train(cfg,
                              model) is nn.parallel.DistributedDataParallel else model.state_dict(),
                          'optimizer': optimizer.state_dict()}
 
-            # Save latest checkpoint
-            torch.save(chkpt, latest)
+            # Save last checkpoint
+            torch.save(chkpt, last)
             if opt.bucket:
-                os.system('gsutil cp %s gs://%s' % (latest, opt.bucket))  # upload to bucket
+                os.system('gsutil cp %s gs://%s' % (last, opt.bucket))  # upload to bucket
 
             # Save best checkpoint
             if best_fitness == fitness:
diff --git a/utils/datasets.py b/utils/datasets.py
index 116aff6a..13ab07fc 100755
--- a/utils/datasets.py
+++ b/utils/datasets.py
@@ -181,9 +181,13 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
                 s = [exif_size(Image.open(f)) for f in tqdm(self.img_files, desc='Reading image shapes')]
                 np.savetxt(sp, s, fmt='%g')
 
-            with open(sp, 'r') as f:  # read existing shapefile
-                s = np.array([x.split() for x in f.read().splitlines()], dtype=np.float64)
-                assert len(s) == n, 'Shapefile error. Please delete %s and rerun' % sp  # TODO: auto-delete shapefile
+            try:
+                with open(sp, 'r') as f:  # read existing shapefile
+                    s = np.array([x.split() for x in f.read().splitlines()], dtype=np.float64)
+                    assert len(s) == n, 'Shapefile out of sync'
+            except:
+                os.remove(sp)
+                print('Shapefile deleted: %s. Please rerun again.' % sp)
 
             # Sort by aspect ratio
             ar = s[:, 1] / s[:, 0]  # aspect ratio
diff --git a/utils/gcp.sh b/utils/gcp.sh
index 5f2e49f2..0379a479 100755
--- a/utils/gcp.sh
+++ b/utils/gcp.sh
@@ -35,12 +35,12 @@ git pull https://github.com/ultralytics/yolov3 test  # branch
 # Test Darknet training
 python3 test.py --weights ../darknet/backup/yolov3.backup
 
-# Copy latest.pt TO bucket
-gsutil cp yolov3/weights/latest1gpu.pt gs://ultralytics
+# Copy last.pt TO bucket
+gsutil cp yolov3/weights/last1gpu.pt gs://ultralytics
 
-# Copy latest.pt FROM bucket
-gsutil cp gs://ultralytics/latest.pt yolov3/weights/latest.pt
-wget https://storage.googleapis.com/ultralytics/yolov3/latest_v1_0.pt -O weights/latest_v1_0.pt
+# Copy last.pt FROM bucket
+gsutil cp gs://ultralytics/last.pt yolov3/weights/last.pt
+wget https://storage.googleapis.com/ultralytics/yolov3/last_v1_0.pt -O weights/last_v1_0.pt
 wget https://storage.googleapis.com/ultralytics/yolov3/best_v1_0.pt -O weights/best_v1_0.pt
 
 # Reproduce tutorials
@@ -94,7 +94,7 @@ python3 test.py --data ../supermarket2/supermarket2.data --weights ../darknet/ba
 
 # Debug/Development
 python3 train.py --data data/coco.data --img-size 320 --single-scale --batch-size 64 --accumulate 1 --epochs 1 --evolve --giou
-python3 test.py --weights weights/latest.pt --cfg cfg/yolov3-spp.cfg --img-size 320
+python3 test.py --weights weights/last.pt --cfg cfg/yolov3-spp.cfg --img-size 320
 
 gsutil cp evolve.txt gs://ultralytics
 sudo shutdown