Remove wandb logging

modeling_diffusiondet.py

@@ -10,7 +10,6 @@ import torch.nn.functional as F
 from torchvision import ops
 from torchvision.ops.feature_pyramid_network import FeaturePyramidNetwork
 from transformers import PreTrainedModel
-import wandb
 
 from transformers.utils.backbone_utils import load_backbone
 from .configuration_diffusiondet import DiffusionDetConfig
@@ -48,6 +47,7 @@ def cosine_beta_schedule(timesteps, s=0.008):
     betas = 1 - (alphas_cumprod[1:] / alphas_cumprod[:-1])
     return torch.clip(betas, 0, 0.999)
 
+
 @dataclass
 class DiffusionDetOutput(ModelOutput):
     """
@@ -60,6 +60,7 @@ class DiffusionDetOutput(ModelOutput):
     labels: torch.IntTensor = None
     pred_boxes: torch.FloatTensor = None
 
+
 class DiffusionDet(PreTrainedModel):
     """
     Implement DiffusionDet
@@ -277,15 +278,8 @@ class DiffusionDet(PreTrainedModel):
                 loss_dict[k] *= weight_dict[k]
         loss_dict['loss'] = sum([loss_dict[k] for k in weight_dict.keys()])
 
-        wandb_logs_values = ["loss_ce", "loss_bbox", "loss_giou"]
-
-        if self.training:
-            wandb.log({f'train/{k}': v.detach().cpu().numpy() for k, v in loss_dict.items() if k in wandb_logs_values})
-        else:
-            wandb.log({f'eval/{k}': v.detach().cpu().numpy() for k, v in loss_dict.items() if k in wandb_logs_values})
-
         if not self.training:
-            pred_logits, pred_labels, pred_boxes  = self.ddim_sample(pixel_values, features, images_whwh)
+            pred_logits, pred_labels, pred_boxes = self.ddim_sample(pixel_values, features, images_whwh)
             return DiffusionDetOutput(
                 loss=loss_dict['loss'],
                 loss_dict=loss_dict,