forked from facebookresearch/mmf
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Fix ocr based configs and add LoRRA configs for VizWiz and VQA2
- Loading branch information
Showing
6 changed files
with
276 additions
and
15 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,120 @@ | ||
| includes: | ||
| - common/defaults/configs/tasks/vqa/vizwiz.yml | ||
| # Use soft copy | ||
| task_attributes: | ||
| vqa: | ||
| dataset_attributes: | ||
| vizwiz: | ||
| use_ocr: true | ||
| processors: | ||
| answer_processor: | ||
| type: soft_copy_answer | ||
| params: | ||
| vocab_file: vocabs/answers_vizwiz_7k.txt | ||
| preprocessor: | ||
| type: simple_word | ||
| params: {} | ||
| context_preprocessor: | ||
| type: simple_word | ||
| params: {} | ||
| max_length: 50 | ||
| num_answers: 10 | ||
| model_attributes: | ||
| lorra: &lorra | ||
| model_data_dir: ../data | ||
| metrics: | ||
| - type: vqa_accuracy | ||
| losses: | ||
| - type: logit_bce | ||
| num_context_features: 1 | ||
| context_feature_dim: 300 | ||
| image_feature_dim: 2048 | ||
| context_max_len: 50 | ||
| classifier: | ||
| type: logit | ||
| params: | ||
| img_hidden_dim: 5000 | ||
| text_hidden_dim: 300 | ||
| image_feature_embeddings: | ||
| - modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| normalization: softmax | ||
| transform: | ||
| type: linear | ||
| params: | ||
| out_dim: 1 | ||
| context_feature_embeddings: | ||
| - modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| normalization: sigmoid | ||
| transform: | ||
| type: linear | ||
| params: | ||
| out_dim: 1 | ||
| image_feature_encodings: | ||
| - type: finetune_faster_rcnn_fpn_fc7 | ||
| params: | ||
| bias_file: detectron/fc6/fc7_b.pkl | ||
| weights_file: detectron/fc6/fc7_w.pkl | ||
| - type: default | ||
| params: {} | ||
| context_feature_encodings: | ||
| - type: default | ||
| params: {} | ||
| image_text_modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| # 300 for FastText and 50 for order vectors | ||
| context_dim: 350 | ||
| text_embeddings: | ||
| - type: attention | ||
| params: | ||
| hidden_dim: 1024 | ||
| num_layers: 1 | ||
| conv1_out: 512 | ||
| conv2_out: 2 | ||
| dropout: 0 | ||
| embedding_dim: 300 | ||
| kernel_size: 1 | ||
| padding: 0 | ||
| context_embeddings: | ||
| - type: identity | ||
| params: | ||
| embedding_dim: 350 | ||
| lorra_with_glove: *lorra | ||
| optimizer_attributes: | ||
| params: | ||
| eps: 1.0e-08 | ||
| lr: 0.01 | ||
| weight_decay: 0 | ||
| type: Adamax | ||
| training_parameters: | ||
| clip_norm_mode: all | ||
| clip_gradients: true | ||
| max_grad_l2_norm: 0.25 | ||
| lr_scheduler: true | ||
| lr_steps: | ||
| - 14000 | ||
| lr_ratio: 0.01 | ||
| use_warmup: true | ||
| warmup_factor: 0.2 | ||
| warmup_iterations: 1000 | ||
| max_iterations: 24000 | ||
| batch_size: 128 | ||
| num_workers: 7 | ||
| task_size_proportional_sampling: true | ||
| monitored_metric: vqa_accuracy | ||
| pretrained_mapping: | ||
| text_embeddings: text_embeddings | ||
| image_feature_encoders: image_feature_encoders | ||
| image_feature_embeddings_list: image_feature_embeddings_list | ||
| image_text_multi_modal_combine_layer: image_text_multi_modal_combine_layer | ||
| metric_minimize: false |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,125 @@ | ||
| includes: | ||
| - common/defaults/configs/tasks/vqa/vqa2.yml | ||
| # Use soft copy | ||
| task_attributes: | ||
| vqa: | ||
| dataset_attributes: | ||
| vqa2: | ||
| use_ocr: true | ||
| processors: | ||
| answer_processor: | ||
| type: soft_copy_answer | ||
| params: | ||
| vocab_file: vocabs/answers_vqa.txt | ||
| preprocessor: | ||
| type: simple_word | ||
| params: {} | ||
| context_preprocessor: | ||
| type: simple_word | ||
| params: {} | ||
| max_length: 50 | ||
| num_answers: 10 | ||
| model_attributes: | ||
| lorra: &lorra | ||
| model_data_dir: ../data | ||
| metrics: | ||
| - type: vqa_accuracy | ||
| losses: | ||
| - type: logit_bce | ||
| num_context_features: 1 | ||
| context_feature_dim: 300 | ||
| image_feature_dim: 2048 | ||
| context_max_len: 50 | ||
| classifier: | ||
| type: logit | ||
| params: | ||
| img_hidden_dim: 5000 | ||
| text_hidden_dim: 300 | ||
| image_feature_embeddings: | ||
| - modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| normalization: softmax | ||
| transform: | ||
| type: linear | ||
| params: | ||
| out_dim: 1 | ||
| context_feature_embeddings: | ||
| - modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| normalization: sigmoid | ||
| transform: | ||
| type: linear | ||
| params: | ||
| out_dim: 1 | ||
| image_feature_encodings: | ||
| - type: finetune_faster_rcnn_fpn_fc7 | ||
| params: | ||
| bias_file: detectron/fc6/fc7_b.pkl | ||
| weights_file: detectron/fc6/fc7_w.pkl | ||
| - type: default | ||
| params: {} | ||
| context_feature_encodings: | ||
| - type: default | ||
| params: {} | ||
| image_text_modal_combine: | ||
| type: non_linear_element_multiply | ||
| params: | ||
| dropout: 0 | ||
| hidden_dim: 5000 | ||
| # 300 for FastText and 50 for order vectors | ||
| context_dim: 350 | ||
| text_embeddings: | ||
| - type: attention | ||
| params: | ||
| hidden_dim: 1024 | ||
| num_layers: 1 | ||
| conv1_out: 512 | ||
| conv2_out: 2 | ||
| dropout: 0 | ||
| embedding_dim: 300 | ||
| kernel_size: 1 | ||
| padding: 0 | ||
| context_embeddings: | ||
| - type: identity | ||
| params: | ||
| embedding_dim: 350 | ||
| lorra_with_glove: *lorra | ||
| optimizer_attributes: | ||
| type: Adamax | ||
| params: | ||
| eps: 1.0e-08 | ||
| lr: 0.01 | ||
| weight_decay: 0 | ||
| training_parameters: | ||
| clip_norm_mode: all | ||
| clip_gradients: true | ||
| lr_ratio: 0.1 | ||
| lr_scheduler: true | ||
| lr_steps: | ||
| - 15000 | ||
| - 18000 | ||
| - 20000 | ||
| - 21000 | ||
| max_grad_l2_norm: 0.25 | ||
| max_iterations: 22000 | ||
| use_warmup: true | ||
| warmup_factor: 0.2 | ||
| warmup_iterations: 1000 | ||
| patience: 4000 | ||
| batch_size: 512 | ||
| num_workers: 7 | ||
| task_size_proportional_sampling: true | ||
| monitored_metric: vqa_accuracy | ||
| pretrained_mapping: | ||
| text_embeddings: text_embeddings | ||
| context_embeddings: context_embeddings | ||
| image_feature_encoders: image_feature_encoders | ||
| image_feature_embeddings_list: image_feature_embeddings_list | ||
| image_text_multi_modal_combine_layer: image_text_multi_modal_combine_layer | ||
| metric_minimize: false |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters