YOLOv5 v4.0 Release (#1837)

* Update C3 module * Update C3 module * Update C3 module * Update C3 module * update * update * update * update * update * update * update * update * update * updates * updates * updates * updates * updates * updates * updates * updates * updates * updates * update * update * update * update * updates * updates * updates * updates * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update datasets * update * update * update * update attempt_downlaod() * merge * merge * update * update * update * update * update * update * update * update * update * update * parameterize eps * comments * gs-multiple * update * max_nms implemented * Create one_cycle() function * update * update * update * update * update * update * update * update study.png * update study.png * Update datasets.py
2021-01-04 19:54:09 -08:00
1 parent 0e341c5660
commit 69be8e738f
@@ -106,7 +106,7 @@ def compute_loss(p, targets, model):  # predictions, targets, model
    # Losses
    nt = 0  # number of targets
    no = len(p)  # number of outputs
-    balance = [4.0, 1.0, 0.4] if no == 3 else [4.0, 1.0, 0.4, 0.1]  # P3-5 or P3-6
+    balance = [4.0, 1.0, 0.3, 0.1, 0.03]  # P3-P7
    for i, pi in enumerate(p):  # layer index, layer predictions
        b, a, gj, gi = indices[i]  # image, anchor, gridy, gridx
        tobj = torch.zeros_like(pi[..., 0], device=device)  # target obj
@@ -140,7 +140,7 @@ def compute_loss(p, targets, model):  # predictions, targets, model

    s = 3 / no  # output count scaling
    lbox *= h['box'] * s
-    lobj *= h['obj'] * s * (1.4 if no == 4 else 1.)
+    lobj *= h['obj']
    lcls *= h['cls'] * s
    bs = tobj.shape[0]  # batch size