Internal change

PiperOrigin-RevId: 364620438

official/vision/beta/configs/experiments/retinanet/coco_spinenet49_mobile_tpu.yaml

+runtime:
+  distribution_strategy: 'tpu'
+  mixed_precision_dtype: 'bfloat16'
+task:
+  losses:
+    l2_weight_decay: 3.0e-05
+  model:
+    anchor:
+      anchor_size: 3
+      aspect_ratios: [0.5, 1.0, 2.0]
+      num_scales: 3
+    backbone:
+      spinenet_mobile:
+        stochastic_depth_drop_rate: 0.2
+        model_id: '49'
+        se_ratio: 0.2
+      type: 'spinenet_mobile'
+    decoder:
+      type: 'identity'
+    head:
+      num_convs: 4
+      num_filters: 48
+      use_separable_conv: true
+    input_size: [384, 384, 3]
+    max_level: 7
+    min_level: 3
+    norm_activation:
+      activation: 'swish'
+      norm_epsilon: 0.001
+      norm_momentum: 0.99
+      use_sync_bn: true
+  train_data:
+    dtype: 'bfloat16'
+    global_batch_size: 256
+    is_training: true
+    parser:
+      aug_rand_hflip: true
+      aug_scale_max: 2.0
+      aug_scale_min: 0.1
+  validation_data:
+    dtype: 'bfloat16'
+    global_batch_size: 8
+    is_training: false
+trainer:
+  checkpoint_interval: 462
+  optimizer_config:
+    learning_rate:
+      stepwise:
+        boundaries: [265650, 272580]
+        values: [0.32, 0.032, 0.0032]
+      type: 'stepwise'
+    warmup:
+      linear:
+        warmup_learning_rate: 0.0067
+        warmup_steps: 2000
+  steps_per_loop: 462
+  train_steps: 277200
+  validation_interval: 462
+  validation_steps: 625

official/vision/beta/configs/experiments/retinanet/coco_spinenet49s_mobile_tpu.yaml

+runtime:
+  distribution_strategy: 'tpu'
+  mixed_precision_dtype: 'bfloat16'
+task:
+  losses:
+    l2_weight_decay: 3.0e-05
+  model:
+    anchor:
+      anchor_size: 3
+      aspect_ratios: [0.5, 1.0, 2.0]
+      num_scales: 3
+    backbone:
+      spinenet_mobile:
+        stochastic_depth_drop_rate: 0.2
+        model_id: '49S'
+        se_ratio: 0.2
+      type: 'spinenet_mobile'
+    decoder:
+      type: 'identity'
+    head:
+      num_convs: 4
+      num_filters: 40
+      use_separable_conv: true
+    input_size: [384, 384, 3]
+    max_level: 7
+    min_level: 3
+    norm_activation:
+      activation: 'swish'
+      norm_epsilon: 0.001
+      norm_momentum: 0.99
+      use_sync_bn: true
+  train_data:
+    dtype: 'bfloat16'
+    global_batch_size: 256
+    is_training: true
+    parser:
+      aug_rand_hflip: true
+      aug_scale_max: 2.0
+      aug_scale_min: 0.1
+  validation_data:
+    dtype: 'bfloat16'
+    global_batch_size: 8
+    is_training: false
+trainer:
+  checkpoint_interval: 462
+  optimizer_config:
+    learning_rate:
+      stepwise:
+        boundaries: [265650, 272580]
+        values: [0.32, 0.032, 0.0032]
+      type: 'stepwise'
+    warmup:
+      linear:
+        warmup_learning_rate: 0.0067
+        warmup_steps: 2000
+  steps_per_loop: 462
+  train_steps: 277200
+  validation_interval: 462
+  validation_steps: 625

official/vision/beta/configs/experiments/retinanet/coco_spinenet49xs_mobile_tpu.yaml

+runtime:
+  distribution_strategy: 'tpu'
+  mixed_precision_dtype: 'bfloat16'
+task:
+  losses:
+    l2_weight_decay: 3.0e-05
+  model:
+    anchor:
+      anchor_size: 3
+      aspect_ratios: [0.5, 1.0, 2.0]
+      num_scales: 3
+    backbone:
+      spinenet_mobile:
+        stochastic_depth_drop_rate: 0.2
+        model_id: '49XS'
+        se_ratio: 0.2
+      type: 'spinenet_mobile'
+    decoder:
+      type: 'identity'
+    head:
+      num_convs: 4
+      num_filters: 24
+      use_separable_conv: false
+    input_size: [256, 256, 3]
+    max_level: 7
+    min_level: 3
+    norm_activation:
+      activation: 'swish'
+      norm_epsilon: 0.001
+      norm_momentum: 0.99
+      use_sync_bn: true
+  train_data:
+    dtype: 'bfloat16'
+    global_batch_size: 256
+    is_training: true
+    parser:
+      aug_rand_hflip: true
+      aug_scale_max: 2.0
+      aug_scale_min: 0.1
+  validation_data:
+    dtype: 'bfloat16'
+    global_batch_size: 8
+    is_training: false
+trainer:
+  checkpoint_interval: 462
+  optimizer_config:
+    learning_rate:
+      stepwise:
+        boundaries: [265650, 272580]
+        values: [0.32, 0.032, 0.0032]
+      type: 'stepwise'
+    warmup:
+      linear:
+        warmup_learning_rate: 0.0067
+        warmup_steps: 2000
+  steps_per_loop: 462
+  train_steps: 277200
+  validation_interval: 462
+  validation_steps: 625

official/vision/beta/configs/retinanet.py

@@ -322,6 +322,7 @@ def retinanet_spinenet_mobile_coco() -> cfg.ExperimentConfig:
                       model_id='49', stochastic_depth_drop_rate=0.2)),
               decoder=decoders.Decoder(
                   type='identity', identity=decoders.Identity()),
+              head=RetinaNetHead(num_filters=48, use_separable_conv=True),
               anchor=Anchor(anchor_size=3),
               norm_activation=common.NormActivation(
                   use_sync_bn=True, activation='swish'),
@@ -329,7 +330,7 @@ def retinanet_spinenet_mobile_coco() -> cfg.ExperimentConfig:
               input_size=[input_size, input_size, 3],
               min_level=3,
               max_level=7),
-          losses=Losses(l2_weight_decay=4e-5),
+          losses=Losses(l2_weight_decay=3e-5),
           train_data=DataConfig(
               input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
               is_training=True,