format

Project-MONAI · Oct 29, 2024 · 0427288 · 0427288
1 parent 385a359
commit 0427288
Show file tree

Hide file tree

Showing 8 changed files with 50 additions and 91 deletions.
diff --git a/m3/demo/gradio_m3.py b/m3/demo/gradio_m3.py
@@ -676,6 +676,7 @@ def update_sys_message(sys_message, sv):
     sv.sys_msg = sys_message
     return sv
 
+
 def update_modality_prompt(modality_prompt, sv):
     """Update the modality prompt"""
     logger.debug(f"Updating the modality prompt")
@@ -785,7 +786,16 @@ def create_demo(source, model_path, conv_mode, server_port):
         clear_btn.click(
             fn=clear_all_convs,
             inputs=[sv],
-            outputs=[sv, prompt_edit, chat_history, history_text, history_text_full, sys_prompt_text, model_cards_text, modality_prompt_dropdown],
+            outputs=[
+                sv,
+                prompt_edit,
+                chat_history,
+                history_text,
+                history_text_full,
+                sys_prompt_text,
+                model_cards_text,
+                modality_prompt_dropdown,
+            ],
         )
 
         # States

diff --git a/m3/eval/scripts/classification/prompts.py b/m3/eval/scripts/classification/prompts.py
@@ -20,8 +20,7 @@
     "incorporate the additional results generated by an expert classification model:\n"
 )
 text_only_expert_prefix = (
-    "When answering the question, please "
-    "incorporate the results generated by an expert classification model:\n"
+    "When answering the question, please incorporate the results generated by an expert classification model:\n"
 )
 multi_choice_prompt = [
     [
@@ -98,7 +97,7 @@ def has_placeholder(prompts, placeholder):
     return False
 
 
-_first_prompt = "Question: is there <class_name> according to the image?\n" "Please reply with yes or no.\n"
+_first_prompt = "Question: is there <class_name> according to the image?\n Please reply with yes or no.\n"
 binary_conv_prompt = [
     [0, model_list + f"<image> This is a CXR image.\n{_first_prompt}"],
     [1, "This looks like a chest x-ray. Let me first trigger <CXR()>."],
@@ -110,4 +109,9 @@ def has_placeholder(prompts, placeholder):
     [1, None],
 ]
 
-templates = {"multi_choice": multi_choice_prompt, "binary": binary_prompt, "binary_conv": binary_conv_prompt, "text_only": text_only_prompt}
+templates = {
+    "multi_choice": multi_choice_prompt,
+    "binary": binary_prompt,
+    "binary_conv": binary_conv_prompt,
+    "text_only": text_only_prompt,
+}
diff --git a/m3/eval/scripts/print_metrics.py b/m3/eval/scripts/print_metrics.py
@@ -71,7 +71,6 @@ def main():
     except:
         results["report_mimiccxr_old_green"] = "?"
 
-
     try:
         json_data = load_json(os.path.join(args.input, "report_mimiccxr/results_clean.json"))
         results["report_mimiccxr_clean_bleu4"] = json_data["BLEU_4"]
@@ -95,7 +94,6 @@ def main():
     except:
         results["report_mimiccxr_clean_expert_bleu4"] = "?"
         results["report_mimiccxr_clean_expert_rougel"] = "?"
-
 
     try:
         json_data = load_json(os.path.join(args.input, "report_mimiccxr/result_clean_expert_green.json"))

diff --git a/m3/eval/scripts/report_updated/cleaning.py b/m3/eval/scripts/report_updated/cleaning.py
@@ -101,20 +101,15 @@ def capitalize_first_letter(sentences):
             # Capitalize the first letter
             refined_sentences.append(sentence[0].upper() + sentence[1:])
         else:
-            print(
-                f"First character is not a letter in sentence: '{sentence}'"
-            )
+            print(f"First character is not a letter in sentence: '{sentence}'")
             continue
 
     return refined_sentences
 
 
 def remove_before_colon(sentences):
     # Replace only when a colon is followed by a space
-    refined_sentences = [
-        sentence.split(": ", 1)[-1] if ": " in sentence else sentence
-        for sentence in sentences
-    ]
+    refined_sentences = [sentence.split(": ", 1)[-1] if ": " in sentence else sentence for sentence in sentences]
 
     return refined_sentences
 
@@ -166,9 +161,7 @@ def remove_duplicate_sentences(sentences):
 
     for sentence in sentences:
         if sentence not in seen:
-            unique_sentences.append(
-                sentence
-            )  # Add sentence to the result if not seen before
+            unique_sentences.append(sentence)  # Add sentence to the result if not seen before
             seen.add(sentence)  # Mark this sentence as seen
 
     return unique_sentences
@@ -209,11 +202,7 @@ def replace_appearance(text, target_word, replacement_word, nth_app):
         return text
 
     # Rebuild the string: join the parts, but replace the second occurrence
-    modified_text = (
-        target_word.join(parts[:nth_app])
-        + replacement_word
-        + target_word.join(parts[nth_app:])
-    )
+    modified_text = target_word.join(parts[:nth_app]) + replacement_word + target_word.join(parts[nth_app:])
 
     return modified_text
 

diff --git a/m3/eval/scripts/report_updated/inference.py b/m3/eval/scripts/report_updated/inference.py
@@ -10,16 +10,14 @@
 # limitations under the License.
 
 import argparse
-import re
-from io import BytesIO
+import math
 import os
 import os.path as osp
+import re
+from io import BytesIO
 
-import math
 import requests
 import torch
-from PIL import Image
-
 from llava.constants import (
     DEFAULT_IM_END_TOKEN,
     DEFAULT_IM_START_TOKEN,
@@ -28,14 +26,10 @@
     IMAGE_TOKEN_INDEX,
 )
 from llava.conversation import SeparatorStyle, conv_templates
-from llava.mm_utils import (
-    KeywordsStoppingCriteria,
-    get_model_name_from_path,
-    process_images,
-    tokenizer_image_token,
-)
+from llava.mm_utils import KeywordsStoppingCriteria, get_model_name_from_path, process_images, tokenizer_image_token
 from llava.model.builder import load_pretrained_model
 from llava.utils import disable_torch_init
+from PIL import Image
 
 
 def load_filenames(file_path):
@@ -80,29 +74,23 @@ def eval_model(args):
     output_folder = args.output_folder
 
     model_name = get_model_name_from_path(args.model_path)
-    tokenizer, model, image_processor, context_len = load_pretrained_model(
-        args.model_path, model_name, args.model_base
-    )
+    tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, model_name, args.model_base)
 
     for img_filename in image_filenames:
         image_path = osp.join(images_folder, img_filename)
         image = load_image(image_path)
 
         query = "Describe the image in detail."
 
-        image_token_se = (
-            DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN
-        )
+        image_token_se = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN
         if IMAGE_PLACEHOLDER in query:
             if model.config.mm_use_im_start_end:
                 query = re.sub(IMAGE_PLACEHOLDER, image_token_se, query)
             else:
                 query = re.sub(IMAGE_PLACEHOLDER, DEFAULT_IMAGE_TOKEN, query)
         else:
             if DEFAULT_IMAGE_TOKEN not in query:
-                print(
-                    f"no <image> tag found in input. Automatically append one at the beginning of text."
-                )
+                print(f"no <image> tag found in input. Automatically append one at the beginning of text.")
                 if model.config.mm_use_im_start_end:
                     query = image_token_se + "\n" + query
                 else:
@@ -133,16 +121,8 @@ def eval_model(args):
         conv.append_message(conv.roles[1], None)
         prompt = conv.get_prompt()
 
-        images_tensor = process_images([image], image_processor, model.config).to(
-            model.device, dtype=torch.float16
-        )
-        input_ids = (
-            tokenizer_image_token(
-                prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
-            )
-            .unsqueeze(0)
-            .cuda()
-        )
+        images_tensor = process_images([image], image_processor, model.config).to(model.device, dtype=torch.float16)
+        input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).cuda()
 
         stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
         keywords = [stop_str]

diff --git a/m3/eval/scripts/report_updated/inference_expert.py b/m3/eval/scripts/report_updated/inference_expert.py
@@ -10,17 +10,15 @@
 # limitations under the License.
 
 import argparse
-import re
-from io import BytesIO
+import json
+import math
 import os
 import os.path as osp
+import re
+from io import BytesIO
 
-import json
-import math
 import requests
 import torch
-from PIL import Image
-
 from llava.constants import (
     DEFAULT_IM_END_TOKEN,
     DEFAULT_IM_START_TOKEN,
@@ -29,15 +27,10 @@
     IMAGE_TOKEN_INDEX,
 )
 from llava.conversation import SeparatorStyle, conv_templates
-from llava.mm_utils import (
-    KeywordsStoppingCriteria,
-    get_model_name_from_path,
-    process_images,
-    tokenizer_image_token,
-)
+from llava.mm_utils import KeywordsStoppingCriteria, get_model_name_from_path, process_images, tokenizer_image_token
 from llava.model.builder import load_pretrained_model
 from llava.utils import disable_torch_init
-
+from PIL import Image
 
 model_list = (
     "Here is a list of available expert models:\n"
@@ -108,9 +101,7 @@ def eval_model(args):
     output_folder = args.output_folder
 
     model_name = get_model_name_from_path(args.model_path)
-    tokenizer, model, image_processor, context_len = load_pretrained_model(
-        args.model_path, model_name, args.model_base
-    )
+    tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, model_name, args.model_base)
 
     for img_filename in image_filenames:
         image_path = osp.join(images_folder, img_filename)
@@ -146,22 +137,14 @@ def eval_model(args):
         conv.append_message(
             conv.roles[0],
             f"The resulting predictions are:\n{preds}. Analyze the image and take these predictions "
-            f"into account when responding to this prompt.\n{query}"
+            f"into account when responding to this prompt.\n{query}",
         )
         conv.append_message(conv.roles[1], None)
         print(conv)
         prompt = conv.get_prompt()
 
-        images_tensor = process_images([image], image_processor, model.config).to(
-            model.device, dtype=torch.float16
-        )
-        input_ids = (
-            tokenizer_image_token(
-                prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
-            )
-            .unsqueeze(0)
-            .cuda()
-        )
+        images_tensor = process_images([image], image_processor, model.config).to(model.device, dtype=torch.float16)
+        input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).cuda()
 
         stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
         keywords = [stop_str]

diff --git a/m3/eval/scripts/report_updated/metric_green_score.py b/m3/eval/scripts/report_updated/metric_green_score.py
@@ -17,6 +17,7 @@
 
 from green_score import GREEN
 
+
 def load_text_files_from_directory(directory):
     """
     Load the content of all text files from a directory and return it as a list of strings.
@@ -26,15 +27,12 @@ def load_text_files_from_directory(directory):
         filepath = os.path.join(directory, filename)
         if os.path.isfile(filepath) and filename.endswith(".txt"):
             with open(filepath, "r") as file:
-                content = (
-                    file.read().strip()
-                )  # Read the file content and remove any surrounding whitespace
+                content = file.read().strip()  # Read the file content and remove any surrounding whitespace
                 texts.append(content)
     return texts
 
 
 def run_inference(refs, hyps):
-
     # Initialize the GREEN model (assumes correct GPU has been set in the environment)
     model = GREEN(
         model_id_or_path="StanfordAIMI/GREEN-radllama2-7b",
@@ -99,9 +97,7 @@ def read_files(directory):
     refs_dir = sys.argv[1]  # refs directory path
     hyps_dir = sys.argv[2]  # hyps directory path
     num_partitions = int(sys.argv[3])  # Total number of partitions (e.g., total GPUs)
-    partition_index = int(
-        sys.argv[4]
-    )  # Index of this partition (e.g., current GPU/process index)
+    partition_index = int(sys.argv[4])  # Index of this partition (e.g., current GPU/process index)
 
     ground_truths = read_files(refs_dir)
     predictions = read_files(hyps_dir)

diff --git a/m3/eval/scripts/report_updated/metrics.py b/m3/eval/scripts/report_updated/metrics.py
@@ -15,9 +15,9 @@
 import re
 
 from pycocoevalcap.bleu.bleu import Bleu
+from pycocoevalcap.cider.cider import Cider
 from pycocoevalcap.meteor.meteor import Meteor
 from pycocoevalcap.rouge.rouge import Rouge
-from pycocoevalcap.cider.cider import Cider
 from pycocoevalcap.tokenizer.ptbtokenizer import PTBTokenizer
 
 
@@ -60,7 +60,7 @@ def compute_scores(gts, res):
     :param res: Dictionary with the image ids ant their generated captions
     :print: Evaluation score (the mean of the scores of all the instances) for each measure
     """
-    print('tokenization...')
+    print("tokenization...")
     tokenizer = PTBTokenizer()
     gts = tokenizer.tokenize(gts)
     res = tokenizer.tokenize(res)
@@ -122,9 +122,7 @@ def main():
 
             _text = predictions[filename]
             _text = normalize_spaces(_text.replace("\n", " "))
-            prediction_data.append(
-                {"image_id": idx, "caption": normalize_spaces(_text.replace("\n", " "))}
-            )
+            prediction_data.append({"image_id": idx, "caption": normalize_spaces(_text.replace("\n", " "))})
 
     print(f"found {len(prediction_data)} prediction data points.")
 
@@ -141,5 +139,6 @@ def main():
     with open(args.output, "w") as f:
         json.dump(eval_res, f)
 
+
 if __name__ == "__main__":
     main()