search

primepake / wav2lip_288x288

MIT License
524 stars 135 forks source link

What indicator represents the end of training hq_wav2lip_sam_train？ #139

Open mrhanCH opened 2 months ago

mrhanCH commented 2 months ago

Thank you for your excellent work. May I ask what indicator represents when I can manually end the training of hq_wav2lip_sam_train? Or will the training process end automatically?

mrhanCH commented 2 months ago

Step 129 | L1: 0.139 | Vgg: 0.4578 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.415 Step 130 | L1: 0.1386 | Vgg: 0.4565 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.006001, Train: 9.395 Step 131 | L1: 0.1384 | Vgg: 0.4553 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005002, Train: 9.412 Step 132 | L1: 0.1385 | Vgg: 0.4548 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005014, Train: 9.39 Step 133 | L1: 0.1383 | Vgg: 0.454 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.397

This is my training progress, I feel it's not quite normal.

ashishjamarkattel commented 2 months ago

did you train syncnet ?

ashishjamarkattel commented 2 months ago

Step 129 | L1: 0.139 | Vgg: 0.4578 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.415 Step 130 | L1: 0.1386 | Vgg: 0.4565 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.006001, Train: 9.395 Step 131 | L1: 0.1384 | Vgg: 0.4553 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005002, Train: 9.412 Step 132 | L1: 0.1385 | Vgg: 0.4548 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005014, Train: 9.39 Step 133 | L1: 0.1383 | Vgg: 0.454 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.397

This is my training progress, I feel it's not quite normal.

you can check the output for the stop, originall loss should be less less than 0.2. If syncnet is trained well.

mrhanCH commented 2 months ago

Step 129 | L1: 0.139 | Vgg: 0.4578 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.415 Step 130 | L1: 0.1386 | Vgg: 0.4565 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.006001, Train: 9.395 Step 131 | L1: 0.1384 | Vgg: 0.4553 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005002, Train: 9.412 Step 132 | L1: 0.1385 | Vgg: 0.4548 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005014, Train: 9.39 Step 133 | L1: 0.1383 | Vgg: 0.454 | SW: 0.03 | Sync: 0.0 | DW: 0.0 | Percep: 0.0 | Fake: 0.0, Real: 0.0 | Load: 0.005001, Train: 9.397 This is my training progress, I feel it's not quite normal.

you can check the output for the stop, originall loss should be less less than 0.2. If syncnet is trained well.

Syncnet trained to 0.3, which label is the origin loss? L1? thanks