Tidy up for project submission

holylovenia · Nov 27, 2022 · c1b3a82 · c1b3a82
1 parent 0f5f545
commit c1b3a82
Show file tree

Hide file tree

Showing 5 changed files with 18 additions and 15 deletions.
diff --git a/README.md b/README.md
@@ -14,6 +14,9 @@ The demand for multimodal dialogue system has been rising in various domains, em
 ```
 pip install -r requirements.txt
 ```
+- To obtain the preprocessed dialogue data, run `./run_data_preprocessing.sh`
+- To obtain the preprocessed visual embeddings from DETR, run `./run_object_detection_finetuning.sh` and `./run_object_detection_prediction.sh`
+- To obtain the preprocessed visual embeddings from ResNet-50 for the SIMMC 2.1 baselines, download from the [Google Drive](https://drive.google.com/file/d/1jr7r5Yaca80W5n0hizOakTG-F1ns6BGv/view?usp=sharing) provided by the organizers, then place the file under `./preprocessed_data/visual_features/`
 
 ## File Structure
 - git submodules

diff --git a/object_detection_prediction.py b/object_detection_prediction.py
@@ -143,7 +143,7 @@ def collate_fn(example_batch):
         pred_last_hidden_state = torch.index_select(outputs.last_hidden_state.squeeze(), 0, sorted_pred_indices)
         pred_boxes = torch.index_select(outputs.pred_boxes.squeeze(), 0, sorted_pred_indices)
         pred_results[MAPPING["id2scene"][batch["image_id"]] + ".png"] = torch.cat([
-            pred_last_hidden_state.detach().cpu(), pred_boxes.detach().cpu()
+            pred_last_hidden_state.detach().cpu() #, pred_boxes.detach().cpu()
         ], dim=1)
     if not os.path.exists(data_args.preprocessed_data_dir):
         os.makedirs(data_args.preprocessed_data_dir)

diff --git a/run_data_preprocessing.sh b/run_data_preprocessing.sh
@@ -1,11 +1,11 @@
 DATA_FOLDER="/home/holy/datasets/simmc2.1/"
 
-# python simmc2/model/ambiguous_candidates/format_ambiguous_candidates_data.py \
-# 	--simmc_train_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_train.json" \
-# 	--simmc_dev_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_dev.json" \
-# 	--simmc_devtest_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_devtest.json" \
-# 	--scene_json_folder "/home/holy/datasets/simmc2.1/public/" \
-# 	--ambiguous_candidates_save_path "./preprocessed_data/ambiguous_candidates/"
+python simmc2/model/ambiguous_candidates/format_ambiguous_candidates_data.py \
+	--simmc_train_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_train.json" \
+	--simmc_dev_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_dev.json" \
+	--simmc_devtest_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_devtest.json" \
+	--scene_json_folder "/home/holy/datasets/simmc2.1/public/" \
+	--ambiguous_candidates_save_path "./preprocessed_data/ambiguous_candidates/"
 
 # python simmc2/model/ambiguous_candidates/format_coref_candidates_data.py \
 # 	--simmc_train_json "/home/holy/datasets/simmc2.1/simmc2.1_dials_dstc11_train.json" \
@@ -14,6 +14,6 @@ DATA_FOLDER="/home/holy/datasets/simmc2.1/"
 # 	--scene_json_folder "/home/holy/datasets/simmc2.1/public/" \
 # 	--coref_candidates_save_path "./preprocessed_data/coref_candidates|no_mm/"
 
-cd simmc2/model/mm_dst
-./run_preprocess_gpt2.sh "/home/holy/projects/ambiguous-mm-dialogue/preprocessed_data/coref_candidates|no_mm" /home/holy/datasets/simmc2.1
+# cd simmc2/model/mm_dst
+# ./run_preprocess_gpt2.sh "/home/holy/projects/ambiguous-mm-dialogue/preprocessed_data/coref_candidates|no_mm" /home/holy/datasets/simmc2.1
 # ./run_preprocess_gpt2.sh "/home/holy/projects/ambiguous-mm-dialogue/preprocessed_data/coref_candidates|mm" /home/holy/datasets/simmc2.1
diff --git a/run_object_detection_finetuning.sh b/run_object_detection_finetuning.sh
@@ -1,10 +1,10 @@
 CUDA_VISIBLE_DEVICES=1 python object_detection_finetuning.py \
-    --output_dir="./save/test/obj-det" \
-    --cache_dir="./cache/test/obj-det" \
+    --output_dir="./save/obj-det" \
+    --cache_dir="./cache/obj-det" \
     --model_name_or_path="facebook/detr-resnet-50" \
-    --train_dataset_path="./preprocessed_data/coref_candidates/simmc2.1_coref_candidates_dstc11_train.json" \
-    --dev_dataset_path="./preprocessed_data/coref_candidates/simmc2.1_coref_candidates_dstc11_dev.json" \
-    --devtest_dataset_path="./preprocessed_data/coref_candidates/simmc2.1_coref_candidates_dstc11_devtest.json" \
+    --train_dataset_path="./preprocessed_data/ambiguous_candidates/simmc2.1_ambiguous_candidates_dstc11_train.json" \
+    --dev_dataset_path="./preprocessed_data/ambiguous_candidates/simmc2.1_ambiguous_candidates_dstc11_dev.json" \
+    --devtest_dataset_path="./preprocessed_data/ambiguous_candidates/simmc2.1_ambiguous_candidates_dstc11_devtest.json" \
     --per_device_train_batch_size=16 \
     --per_device_eval_batch_size=16 \
     --num_train_epochs=200 \

diff --git a/run_object_detection_prediction.sh b/run_object_detection_prediction.sh
@@ -4,7 +4,7 @@ CUDA_VISIBLE_DEVICES=0 python object_detection_prediction.py \
     --dev_dataset_path='./preprocessed_data/ambiguous_candidates/simmc2.1_ambiguous_candidates_dstc11_dev.json' \
     --devtest_dataset_path='./preprocessed_data/ambiguous_candidates/simmc2.1_ambiguous_candidates_dstc11_devtest.json' \
     --cache_dir="./cache/obj-det" \
-    --model_name_or_path="./save/obj-det/facebook/detr-resnet-50/checkpoint-1160" \
+    --model_name_or_path="./save/obj-det/facebook/detr-resnet-50" \
     --per_device_train_batch_size=1 \
     --per_device_eval_batch_size=1 \
     --num_train_epochs=200 \