wip

supervisely_integration/train/hyperparameters.yaml

@@ -1,34 +1,67 @@
-general:
-  num_epochs: 20 # Total number of training epochs
+HybridEncoder:
+  eval_spatial_size:
+    - 640
+    - 640
+RTDETRTransformer:
+  eval_spatial_size:
+    - 640
+    - 640
 
-data: # Section for data-related parameters
-  input_size: [640, 640] # Input image size (width, height)
-  train_batch_size: 2 # Batch size for training
-  val_batch_size: 1 # Batch size for validation
-  val_interval: 1 # Interval for running validation
-
-checkpoint:
-  interval: 1 # Interval for saving checkpoints
-  keep_checkpoints: true # Whether to keep saved checkpoints
-  max_checkpoints: 3 # Maximum number of checkpoints to retain
-  save_last: true # Save the last checkpoint after training
-  save_best: true # Save the best-performing checkpoint based on validation
-  save_optimizer_state: false # Save the optimizer state in checkpoints
+checkpoint_step: 1
+clip_max_norm: -1
+ema:
+  decay: 0.9999
+  type: ModelEMA
+  warmups: 2000
+epoches: 1
+lr_scheduler: null
+lr_warmup:
+  end_factor: 1.0
+  start_factor: 0.001
+  total_iters: 25
+  type: LinearLR
 
 optimizer:
-  frozen_stages_override: false # Override frozen stages if applicable
-  type: "AdamW" # Type of optimizer to use
-  learning_rate: 0.0001 # Initial learning rate
-  weight_decay: 0.0001 # Weight decay (L2 regularization factor)
-  clip_grad_norm: true # Enable gradient clipping
-  grad_norm_value: 0.1 # Maximum norm for gradient clipping
+  betas:
+    - 0.9
+    - 0.999
+  lr: 0.0002
+  type: Adam
+  weight_decay: 0.0001
+remap_mscoco_category: false
+save_ema: false
+save_optimizer: false
+# general:
+#   num_epochs: 20 # Total number of training epochs
+
+# data: # Section for data-related parameters
+#   input_size: [640, 640] # Input image size (width, height)
+#   train_batch_size: 2 # Batch size for training
+#   val_batch_size: 1 # Batch size for validation
+#   val_interval: 1 # Interval for running validation
+
+# checkpoint:
+#   interval: 1 # Interval for saving checkpoints
+#   keep_checkpoints: true # Whether to keep saved checkpoints
+#   max_checkpoints: 3 # Maximum number of checkpoints to retain
+#   save_last: true # Save the last checkpoint after training
+#   save_best: true # Save the best-performing checkpoint based on validation
+#   save_optimizer_state: false # Save the optimizer state in checkpoints
+
+# optimizer:
+#   frozen_stages_override: false # Override frozen stages if applicable
+#   type: "AdamW" # Type of optimizer to use
+#   learning_rate: 0.0001 # Initial learning rate
+#   weight_decay: 0.0001 # Weight decay (L2 regularization factor)
+#   clip_grad_norm: true # Enable gradient clipping
+#   grad_norm_value: 0.1 # Maximum norm for gradient clipping
 
-lr_scheduler:
-  type: null # Type of learning rate scheduler
-  by_epoch: true # Schedule learning rate by epochs rather than steps
-  warmup:
-    enabled: true # Enable warmup phase for learning rate
-    steps: 1 # Number of warmup steps
-    ratio: 0.001 # Starting learning rate ratio for warmup
-    start_factor: 0.001 # Starting learning rate factor for warmup
-    end_factor: 1.0 # Ending learning rate factor for warmup
+# lr_scheduler:
+#   type: null # Type of learning rate scheduler
+#   by_epoch: true # Schedule learning rate by epochs rather than steps
+#   warmup:
+#     enabled: true # Enable warmup phase for learning rate
+#     steps: 1 # Number of warmup steps
+#     ratio: 0.001 # Starting learning rate ratio for warmup
+#     start_factor: 0.001 # Starting learning rate factor for warmup
+#     end_factor: 1.0 # Ending learning rate factor for warmup

supervisely_integration/train/main.py

@@ -30,6 +30,7 @@
 default_config_path = os.path.join(config_paths_dir, "placeholder.yml")
 
 app_options = {
+    "use_coco_annotation": True,
     "save_best_checkpoint": True,
     "save_last_checkpoint": True,
     "supported_train_modes": ["finetune", "scratch"],
@@ -74,14 +75,22 @@
 
 current_file_dir = os.path.dirname(os.path.abspath(__file__))
 output_dir = os.path.join(current_file_dir, "output")
-train = TrainApp(output_dir, models_path, hyperparameters_path, app_options)
+train = TrainApp(models_path, hyperparameters_path, app_options, output_dir)
+
+train
 
 
 @train.start
 def start_training():
     print("-----------------")
     print("Start training")
     print("-----------------")
+
+    # Step 1. convert to COCO format
+    # sly.xxx.convert_to_coco()
+    # Step 2. prepare config.yml (hyperparameters + custom config)
+    # Step 3. train
+
     import rtdetr_pytorch.train as train_cli
 
     custom_config_path = os.path.join(config_paths_dir, "custom.yml")