From b649a95c9a1d334c5ea7b9c86ed2f2cd60a4de1d Mon Sep 17 00:00:00 2001
From: glenn-jocher <glenn.jocher@ultralytics.com>
Date: Fri, 5 Jul 2019 00:36:37 +0200
Subject: [PATCH] GIoU to default

---
 test.py        |  3 +--
 train.py       |  4 ++--
 utils/utils.py | 10 +++++-----
 3 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/test.py b/test.py
index eeaadf6f..f6408699 100644
--- a/test.py
+++ b/test.py
@@ -71,8 +71,7 @@ def test(
 
         # Compute loss
         if hasattr(model, 'hyp'):  # if model has loss hyperparameters
-            loss_i, _ = compute_loss(train_out, targets, model)
-            loss += loss_i.item()
+            loss += compute_loss(train_out, targets, model)[0].item()
 
         # Run NMS
         output = non_max_suppression(inf_out, conf_thres=conf_thres, nms_thres=nms_thres)
diff --git a/train.py b/train.py
index 13cc0d47..2db13eec 100644
--- a/train.py
+++ b/train.py
@@ -218,7 +218,7 @@ def train(
             pred = model(imgs)
 
             # Compute loss
-            loss, loss_items = compute_loss(pred, targets, model, giou_loss=opt.giou)
+            loss, loss_items = compute_loss(pred, targets, model, giou_loss=not opt.xywh)
             if torch.isnan(loss):
                 print('WARNING: nan loss detected, ending training')
                 return results
@@ -320,7 +320,7 @@ if __name__ == '__main__':
     parser.add_argument('--num-workers', type=int, default=4, help='number of Pytorch DataLoader workers')
     parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
     parser.add_argument('--notest', action='store_true', help='only test final epoch')
-    parser.add_argument('--giou', action='store_true', help='use GIoU loss instead of xy, wh loss')
+    parser.add_argument('--xywh', action='store_true', help='use xywh loss instead of GIoU loss')
     parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters')
     parser.add_argument('--cloud-evolve', action='store_true', help='evolve hyperparameters from a cloud source')
     parser.add_argument('--var', default=0, type=int, help='debug variable')
diff --git a/utils/utils.py b/utils/utils.py
index 40ab7c68..767fd21e 100755
--- a/utils/utils.py
+++ b/utils/utils.py
@@ -271,7 +271,7 @@ def wh_iou(box1, box2):
     return inter_area / union_area  # iou
 
 
-def compute_loss(p, targets, model, giou_loss=False):  # predictions, targets, model
+def compute_loss(p, targets, model, giou_loss=True):  # predictions, targets, model
     ft = torch.cuda.FloatTensor if p[0].is_cuda else torch.Tensor
     lxy, lwh, lcls, lobj = ft([0]), ft([0]), ft([0]), ft([0])
     txy, twh, tcls, tbox, indices, anchor_vec = build_targets(model, targets)
@@ -336,17 +336,17 @@ def build_targets(model, targets):
         if nt:
             iou = torch.stack([wh_iou(x, gwh) for x in layer.anchor_vec], 0)
 
-            use_best = True
-            if use_best:
+            use_best_anchor = False
+            if use_best_anchor:
                 iou, a = iou.max(0)  # best iou and anchor
-            else:
+            else:  # use all anchors
                 na = len(layer.anchor_vec)  # number of anchors
                 a = torch.arange(na).view((-1, 1)).repeat([1, nt]).view(-1)
                 t = targets.repeat([na, 1])
                 gwh = gwh.repeat([na, 1])
                 iou = iou.view(-1)  # use all ious
 
-            # reject below threshold ious (OPTIONAL, increases P, lowers R)
+            # reject anchors below iou_thres (OPTIONAL, increases P, lowers R)
             reject = True
             if reject:
                 j = iou > iou_thres