hotfix(eval): updating num_frames into eval.

Author: Kin-Zhang

eval.py

@@ -48,6 +48,7 @@ def main(cfg):
     checkpoint_params = DictConfig(torch_load_ckpt["hyper_parameters"])
     cfg.output = checkpoint_params.cfg.output + f"-e{torch_load_ckpt['epoch']}-{cfg.av2_mode}-v{cfg.leaderboard_version}"
     cfg.model.update(checkpoint_params.cfg.model)
+    cfg.num_frames = cfg.model.target.get('num_frames', checkpoint_params.cfg.get('num_frames', cfg.get('num_frames', 2)))
 
     mymodel = ModelWrapper.load_from_checkpoint(cfg.checkpoint, cfg=cfg, eval=True)
     print(f"\n---LOG[eval]: Loaded model from {cfg.checkpoint}. The backbone network is {checkpoint_params.cfg.model.name}.\n")
@@ -63,7 +64,7 @@ def main(cfg):
     trainer.validate(model = mymodel, \
                      dataloaders = DataLoader( \
                                             HDF5Dataset(cfg.dataset_path + f"/{cfg.av2_mode}", \
-                                                        n_frames=checkpoint_params.cfg.num_frames  if 'num_frames' in checkpoint_params.cfg else 2, \
+                                                        n_frames=cfg.num_frames, \
                                                         eval=True, leaderboard_version=cfg.leaderboard_version), \
                                             batch_size=1, shuffle=False))
     wandb.finish()

save.py

@@ -45,6 +45,7 @@ def main(cfg):
     checkpoint_params = DictConfig(torch.load(cfg.checkpoint)["hyper_parameters"])
     cfg.output = checkpoint_params.cfg.output
     cfg.model.update(checkpoint_params.cfg.model)
+    cfg.num_frames = cfg.model.target.get('num_frames', checkpoint_params.cfg.get('num_frames', cfg.get('num_frames', 2)))
     mymodel = ModelWrapper.load_from_checkpoint(cfg.checkpoint, cfg=cfg, eval=True)
 
     wandb_logger = WandbLogger(save_dir=output_dir,
@@ -57,7 +58,7 @@ def main(cfg):
     # NOTE(Qingwen): search & check in pl_model.py : def test_step(self, batch, res_dict)
     trainer.test(model = mymodel, \
                  dataloaders = DataLoader(\
-                     HDF5Dataset(cfg.dataset_path, n_frames=checkpoint_params.cfg.num_frames if 'num_frames' in checkpoint_params.cfg else 2), \
+                     HDF5Dataset(cfg.dataset_path, n_frames=cfg.num_frames), \
                     batch_size=1, shuffle=False))
     wandb.finish()
 

src/trainer.py

@@ -94,6 +94,7 @@ def __init__(self, cfg, eval=False):
             print(f"We are in {cfg.av2_mode}, results will be saved in: {self.save_res_path} with version: {self.leaderboard_version} format for online leaderboard.")
 
         # self.test_total_num = 0
+        print(cfg)
         self.save_hyperparameters()
 
     # FIXME(Qingwen 2025-08-20): update the loss_calculation fn alone to make all things pretty here....
@@ -208,7 +209,7 @@ def on_validation_epoch_end(self):
         self.model.timer.print(random_colors=False, bold=False)
 
         if self.av2_mode == 'test':
-            print(f"\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.load_checkpoint_path}")
+            print(f"\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.checkpoint}")
             print(f"Test results saved in: {self.save_res_path}, Please run submit command and upload to online leaderboard for results.")
             if self.leaderboard_version == 1:
                 print(f"\nevalai challenge 2010 phase 4018 submit --file {self.save_res_path}.zip --large --private\n")
@@ -221,8 +222,8 @@ def on_validation_epoch_end(self):
             return
 
         if self.av2_mode == 'val':
-            print(f"\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.load_checkpoint_path}")
-            print(f"More details parameters and training status are in checkpoints")        
+            print(f"\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.checkpoint}")
+            print(f"More details parameters and training status are in the checkpoint file.")        
 
         self.metrics.normalize()
 
@@ -339,7 +340,7 @@ def test_step(self, batch, batch_idx):
 
     def on_test_epoch_end(self):
         self.model.timer.print(random_colors=False, bold=False)
-        print(f"\n\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.load_checkpoint_path}")
+        print(f"\n\nModel: {self.model.__class__.__name__}, Checkpoint from: {self.checkpoint}")
         print(f"We already write the flow_est into the dataset, please run following commend to visualize the flow. Copy and paste it to your terminal:")
         print(f"python tools/visualization.py --res_name '{self.vis_name}' --data_dir {self.dataset_path}")
         print(f"Enjoy! ^v^ ------ \n")