Replace pluggable components to create a proper Configurable hierarchy.

Summary:
This large diff rewrites a significant portion of Implicitron's config hierarchy. The new hierarchy, and some of the default implementation classes, are as follows:
```
Experiment
    data_source: ImplicitronDataSource
        dataset_map_provider
        data_loader_map_provider
    model_factory: ImplicitronModelFactory
        model: GenericModel
    optimizer_factory: ImplicitronOptimizerFactory
    training_loop: ImplicitronTrainingLoop
        evaluator: ImplicitronEvaluator
```

1) Experiment (used to be ExperimentConfig) is now a top-level Configurable and contains as members mainly (mostly new) high-level factory Configurables.
2) Experiment's job is to run factories, do some accelerate setup and then pass the results to the main training loop.
3) ImplicitronOptimizerFactory and ImplicitronModelFactory are new high-level factories that create the optimizer, scheduler, model, and stats objects.
4) TrainingLoop is a new configurable that runs the main training loop and the inner train-validate step.
5) Evaluator is a new configurable that TrainingLoop uses to run validation/test steps.
6) GenericModel is not the only model choice anymore. Instead, ImplicitronModelBase (by default instantiated with GenericModel) is a member of Experiment and can be easily replaced by a custom implementation by the user.

All the new Configurables are children of ReplaceableBase, and can be easily replaced with custom implementations.

In addition, I added support for the exponential LR schedule, updated the config files and the test, as well as added a config file that reproduces NERF results and a test to run the repro experiment.

Reviewed By: bottler

Differential Revision: D37723227

fbshipit-source-id: b36bee880d6aa53efdd2abfaae4489d8ab1e8a27

projects/implicitron_trainer/configs/repro_base.yaml

@@ -2,10 +2,10 @@ defaults:
 - default_config
 - _self_
 exp_dir: ./data/exps/base/
-architecture: generic
-visualize_interval: 0
-visdom_port: 8097
-data_source_args:
+training_loop_ImplicitronTrainingLoop_args:
+  visualize_interval: 0
+  max_epochs: 1000
+data_source_ImplicitronDataSource_args:
   data_loader_map_provider_class_type: SequenceDataLoaderMapProvider
   dataset_map_provider_class_type: JsonIndexDatasetMapProvider
   data_loader_map_provider_SequenceDataLoaderMapProvider_args:
@@ -21,55 +21,61 @@ data_source_args:
       load_point_clouds: false
       mask_depths: false
       mask_images: false
-generic_model_args:
-  loss_weights:
-    loss_mask_bce: 1.0
-    loss_prev_stage_mask_bce: 1.0
-    loss_autodecoder_norm: 0.01
-    loss_rgb_mse: 1.0
-    loss_prev_stage_rgb_mse: 1.0
-  output_rasterized_mc: false
-  chunk_size_grid: 102400
-  render_image_height: 400
-  render_image_width: 400
-  num_passes: 2
-  implicit_function_NeuralRadianceFieldImplicitFunction_args:
-    n_harmonic_functions_xyz: 10
-    n_harmonic_functions_dir: 4
-    n_hidden_neurons_xyz: 256
-    n_hidden_neurons_dir: 128
-    n_layers_xyz: 8
-    append_xyz:
-    - 5
-    latent_dim: 0
-  raysampler_AdaptiveRaySampler_args:
-    n_rays_per_image_sampled_from_mask: 1024
-    scene_extent: 8.0
-    n_pts_per_ray_training: 64
-    n_pts_per_ray_evaluation: 64
-    stratified_point_sampling_training: true
-    stratified_point_sampling_evaluation: false
-  renderer_MultiPassEmissionAbsorptionRenderer_args:
-    n_pts_per_ray_fine_training: 64
-    n_pts_per_ray_fine_evaluation: 64
-    append_coarse_samples_to_fine: true
-    density_noise_std_train: 1.0
-  view_pooler_args:
-    view_sampler_args:
-      masked_sampling: false
-  image_feature_extractor_ResNetFeatureExtractor_args:
-    stages:
-    - 1
-    - 2
-    - 3
-    - 4
-    proj_dim: 16
-    image_rescale: 0.32
-    first_max_pool: false
-solver_args:
-  breed: adam
+model_factory_ImplicitronModelFactory_args:
+  visdom_port: 8097
+  model_GenericModel_args:
+    loss_weights:
+      loss_mask_bce: 1.0
+      loss_prev_stage_mask_bce: 1.0
+      loss_autodecoder_norm: 0.01
+      loss_rgb_mse: 1.0
+      loss_prev_stage_rgb_mse: 1.0
+    output_rasterized_mc: false
+    chunk_size_grid: 102400
+    render_image_height: 400
+    render_image_width: 400
+    num_passes: 2
+    implicit_function_NeuralRadianceFieldImplicitFunction_args:
+      n_harmonic_functions_xyz: 10
+      n_harmonic_functions_dir: 4
+      n_hidden_neurons_xyz: 256
+      n_hidden_neurons_dir: 128
+      n_layers_xyz: 8
+      append_xyz:
+      - 5
+      latent_dim: 0
+    raysampler_AdaptiveRaySampler_args:
+      n_rays_per_image_sampled_from_mask: 1024
+      scene_extent: 8.0
+      n_pts_per_ray_training: 64
+      n_pts_per_ray_evaluation: 64
+      stratified_point_sampling_training: true
+      stratified_point_sampling_evaluation: false
+    renderer_MultiPassEmissionAbsorptionRenderer_args:
+      n_pts_per_ray_fine_training: 64
+      n_pts_per_ray_fine_evaluation: 64
+      append_coarse_samples_to_fine: true
+      density_noise_std_train: 1.0
+    view_pooler_args:
+      view_sampler_args:
+        masked_sampling: false
+    image_feature_extractor_ResNetFeatureExtractor_args:
+      stages:
+      - 1
+      - 2
+      - 3
+      - 4
+      proj_dim: 16
+      image_rescale: 0.32
+      first_max_pool: false
+optimizer_factory_ImplicitronOptimizerFactory_args:
+  breed: Adam
+  weight_decay: 0.0
+  lr_policy: MultiStepLR
+  multistep_lr_milestones: []
   lr: 0.0005
-  lr_policy: multistep
-  max_epochs: 2000
+  gamma: 0.1
   momentum: 0.9
-  weight_decay: 0.0
+  betas:
+  - 0.9
+  - 0.999

projects/implicitron_trainer/configs/repro_feat_extractor_normed.yaml

@@ -1,17 +1,18 @@
-generic_model_args:
-  image_feature_extractor_class_type: ResNetFeatureExtractor
-  image_feature_extractor_ResNetFeatureExtractor_args:
-    add_images: true
-    add_masks: true
-    first_max_pool: true
-    image_rescale: 0.375
-    l2_norm: true
-    name: resnet34
-    normalize_image: true
-    pretrained: true
-    stages:
-    - 1
-    - 2
-    - 3
-    - 4
-    proj_dim: 32
+model_factory_ImplicitronModelFactory_args:
+  model_GenericModel_args:
+    image_feature_extractor_class_type: ResNetFeatureExtractor
+    image_feature_extractor_ResNetFeatureExtractor_args:
+      add_images: true
+      add_masks: true
+      first_max_pool: true
+      image_rescale: 0.375
+      l2_norm: true
+      name: resnet34
+      normalize_image: true
+      pretrained: true
+      stages:
+      - 1
+      - 2
+      - 3
+      - 4
+      proj_dim: 32

projects/implicitron_trainer/configs/repro_feat_extractor_transformer.yaml

@@ -1,17 +1,18 @@
-generic_model_args:
-  image_feature_extractor_class_type: ResNetFeatureExtractor
-  image_feature_extractor_ResNetFeatureExtractor_args:
-    add_images: true
-    add_masks: true
-    first_max_pool: false
-    image_rescale: 0.375
-    l2_norm: true
-    name: resnet34
-    normalize_image: true
-    pretrained: true
-    stages:
-    - 1
-    - 2
-    - 3
-    - 4
-    proj_dim: 16
+model_factory_ImplicitronModelFactory_args:
+  model_GenericModel_args:
+    image_feature_extractor_class_type: ResNetFeatureExtractor
+    image_feature_extractor_ResNetFeatureExtractor_args:
+      add_images: true
+      add_masks: true
+      first_max_pool: false
+      image_rescale: 0.375
+      l2_norm: true
+      name: resnet34
+      normalize_image: true
+      pretrained: true
+      stages:
+      - 1
+      - 2
+      - 3
+      - 4
+      proj_dim: 16

projects/implicitron_trainer/configs/repro_feat_extractor_unnormed.yaml

@@ -1,18 +1,19 @@
-generic_model_args:
-  image_feature_extractor_class_type: ResNetFeatureExtractor
-  image_feature_extractor_ResNetFeatureExtractor_args:
-    stages:
-    - 1
-    - 2
-    - 3
-    first_max_pool: false
-    proj_dim: -1
-    l2_norm: false
-    image_rescale: 0.375
-    name: resnet34
-    normalize_image: true
-    pretrained: true
-  view_pooler_args:
-    feature_aggregator_AngleWeightedReductionFeatureAggregator_args:
-      reduction_functions:
-      - AVG
+model_factory_ImplicitronModelFactory_args:
+  model_GenericModel_args:
+    image_feature_extractor_class_type: ResNetFeatureExtractor
+    image_feature_extractor_ResNetFeatureExtractor_args:
+      stages:
+      - 1
+      - 2
+      - 3
+      first_max_pool: false
+      proj_dim: -1
+      l2_norm: false
+      image_rescale: 0.375
+      name: resnet34
+      normalize_image: true
+      pretrained: true
+    view_pooler_args:
+      feature_aggregator_AngleWeightedReductionFeatureAggregator_args:
+        reduction_functions:
+        - AVG

projects/implicitron_trainer/configs/repro_multiseq_base.yaml

@@ -1,7 +1,7 @@
 defaults:
 - repro_base.yaml
 - _self_
-data_source_args:
+data_source_ImplicitronDataSource_args:
   data_loader_map_provider_SequenceDataLoaderMapProvider_args:
     batch_size: 10
     dataset_length_train: 1000
@@ -26,10 +26,12 @@ data_source_args:
     n_frames_per_sequence: -1
     test_on_train: true
     test_restrict_sequence_id: 0
-solver_args:
-  max_epochs: 3000
-  milestones:
+optimizer_factory_ImplicitronOptimizerFactory_args:
+  multistep_lr_milestones:
   - 1000
-camera_difficulty_bin_breaks:
-  - 0.666667
-  - 0.833334
+training_loop_ImplicitronTrainingLoop_args:
+  max_epochs: 3000
+  evaluator_ImplicitronEvaluator_args:
+    camera_difficulty_bin_breaks:
+      - 0.666667
+      - 0.833334

projects/implicitron_trainer/configs/repro_multiseq_idr_ad.yaml

@@ -1,65 +1,66 @@
 defaults:
 - repro_multiseq_base.yaml
 - _self_
-generic_model_args:
-  loss_weights:
-    loss_mask_bce: 100.0
-    loss_kl: 0.0
-    loss_rgb_mse: 1.0
-    loss_eikonal: 0.1
-  chunk_size_grid: 65536
-  num_passes: 1
-  output_rasterized_mc: true
-  sampling_mode_training: mask_sample
-  global_encoder_class_type: SequenceAutodecoder
-  global_encoder_SequenceAutodecoder_args:
-    autodecoder_args:
-      n_instances: 20000
-      init_scale: 1.0
-      encoding_dim: 256
-  implicit_function_IdrFeatureField_args:
-    n_harmonic_functions_xyz: 6
-    bias: 0.6
-    d_in: 3
-    d_out: 1
-    dims:
-    - 512
-    - 512
-    - 512
-    - 512
-    - 512
-    - 512
-    - 512
-    - 512
-    geometric_init: true
-    pooled_feature_dim: 0
-    skip_in:
-    - 6
-    weight_norm: true
-  renderer_SignedDistanceFunctionRenderer_args:
-    ray_tracer_args:
-      line_search_step: 0.5
-      line_step_iters: 3
-      n_secant_steps: 8
-      n_steps: 100
-      object_bounding_sphere: 8.0
-      sdf_threshold: 5.0e-05
-    ray_normal_coloring_network_args:
-      d_in: 9
-      d_out: 3
+model_factory_ImplicitronModelFactory_args:
+  model_GenericModel_args:
+    loss_weights:
+      loss_mask_bce: 100.0
+      loss_kl: 0.0
+      loss_rgb_mse: 1.0
+      loss_eikonal: 0.1
+    chunk_size_grid: 65536
+    num_passes: 1
+    output_rasterized_mc: true
+    sampling_mode_training: mask_sample
+    global_encoder_class_type: SequenceAutodecoder
+    global_encoder_SequenceAutodecoder_args:
+      autodecoder_args:
+        n_instances: 20000
+        init_scale: 1.0
+        encoding_dim: 256
+    implicit_function_IdrFeatureField_args:
+      n_harmonic_functions_xyz: 6
+      bias: 0.6
+      d_in: 3
+      d_out: 1
       dims:
       - 512
       - 512
       - 512
       - 512
-      mode: idr
-      n_harmonic_functions_dir: 4
+      - 512
+      - 512
+      - 512
+      - 512
+      geometric_init: true
       pooled_feature_dim: 0
+      skip_in:
+      - 6
       weight_norm: true
-  raysampler_AdaptiveRaySampler_args:
-    n_rays_per_image_sampled_from_mask: 1024
-    n_pts_per_ray_training: 0
-    n_pts_per_ray_evaluation: 0
-    scene_extent: 8.0
-  renderer_class_type: SignedDistanceFunctionRenderer
-  implicit_function_class_type: IdrFeatureField
+    renderer_SignedDistanceFunctionRenderer_args:
+      ray_tracer_args:
+        line_search_step: 0.5
+        line_step_iters: 3
+        n_secant_steps: 8
+        n_steps: 100
+        object_bounding_sphere: 8.0
+        sdf_threshold: 5.0e-05
+      ray_normal_coloring_network_args:
+        d_in: 9
+        d_out: 3
+        dims:
+        - 512
+        - 512
+        - 512
+        - 512
+        mode: idr
+        n_harmonic_functions_dir: 4
+        pooled_feature_dim: 0
+        weight_norm: true
+    raysampler_AdaptiveRaySampler_args:
+      n_rays_per_image_sampled_from_mask: 1024
+      n_pts_per_ray_training: 0
+      n_pts_per_ray_evaluation: 0
+      scene_extent: 8.0
+    renderer_class_type: SignedDistanceFunctionRenderer
+    implicit_function_class_type: IdrFeatureField