s444415/donut
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

config and params for donut-eval

Browse Source
This commit is contained in:
Michał Kozłowski 2022-12-16 13:53:03 +01:00
parent d98383197f
commit 8ccd1aabb6
2 changed files with 68 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
config.yaml Normal file
View File

@ -0,0 +1,7 @@
pretrained_processor_path: "Zombely/plwiki-proto-fine-tuned-v2"
pretrained_model_path: "Zombely/plwiki-proto-fine-tuned-v2"
validation_dataset_path: "Zombely/diachronia-ocr"
validation_dataset_split: "train"
has_metadata: False
print_output: True
output_file_dir: "../../gonito-outs"

127
donut-eval.py
View File

@ -1,9 +1,6 @@
#!/usr/bin/env python
# coding: utf-8
# In[1]:
from transformers import DonutProcessor, VisionEncoderDecoderModel
from datasets import load_dataset
import re
@ -13,75 +10,71 @@ from tqdm.auto import tqdm
import numpy as np
import pandas as pd
from donut import JSONParseEvaluator
import argparse
from sconf import Config
def main(config):
processor = DonutProcessor.from_pretrained(config.pretrained_processor_path)
model = VisionEncoderDecoderModel.from_pretrained(config.pretrained_model_path)
dataset = load_dataset(config.validation_dataset_path, split=config.validation_dataset_split)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.eval()
model.to(device)
output_list = []
accs = []
# In[2]:
for idx, sample in tqdm(enumerate(dataset), total=len(dataset)):
# prepare encoder inputs
pixel_values = processor(sample['image'].convert("RGB"), return_tensors="pt").pixel_values
pixel_values = pixel_values.to(device)
# prepare decoder inputs
task_prompt = "<s_cord-v2>"
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
decoder_input_ids = decoder_input_ids.to(device)
# autoregressively generate sequence
outputs = model.generate(
pixel_values,
decoder_input_ids=decoder_input_ids,
max_length=model.decoder.config.max_position_embeddings,
early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True,
num_beams=1,
bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True,
)
processor = DonutProcessor.from_pretrained("Zombely/plwiki-proto-fine-tuned-v2")
model = VisionEncoderDecoderModel.from_pretrained("Zombely/plwiki-proto-fine-tuned-v2")
# turn into JSON
seq = processor.batch_decode(outputs.sequences)[0]
seq = seq.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
seq = re.sub(r"<.*?>", "", seq, count=1).strip() # remove first task start token
seq = processor.token2json(seq)
if config.has_metadata:
ground_truth = json.loads(sample["ground_truth"])
ground_truth = ground_truth["gt_parse"]
evaluator = JSONParseEvaluator()
score = evaluator.cal_acc(seq, ground_truth)
accs.append(score)
if config.print_output:
print(seq)
output_list.append(seq)
if config.output_file_dir:
df = pd.DataFrame(map(lambda x: x.get('text_sequence', ''), output_list))
df.to_csv(f'{config.output_file_dir}/{config.pretrained_processor_path}-out.tsv', sep='\t', header=False, index=False)
# In[3]:
if config.has_metadata:
scores = {"accuracies": accs, "mean_accuracy": np.mean(accs)}
print(scores, f"length : {len(accs)}")
print("Mean accuracy:", np.mean(accs))
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--config", type=str, required=True)
args, left_argv = parser.parse_known_args()
config = Config(args.config)
config.argv_update(left_argv)
dataset = load_dataset("Zombely/diachronia-ocr", split='train')
# In[4]:
device = "cuda" if torch.cuda.is_available() else "cpu"
model.eval()
model.to(device)
output_list = []
accs = []
has_metadata = bool(dataset[0].get('ground_truth'))
for idx, sample in tqdm(enumerate(dataset), total=len(dataset)):
# prepare encoder inputs
pixel_values = processor(sample['image'].convert("RGB"), return_tensors="pt").pixel_values
pixel_values = pixel_values.to(device)
# prepare decoder inputs
task_prompt = "<s_cord-v2>"
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
decoder_input_ids = decoder_input_ids.to(device)
# autoregressively generate sequence
outputs = model.generate(
pixel_values,
decoder_input_ids=decoder_input_ids,
max_length=model.decoder.config.max_position_embeddings,
early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True,
num_beams=1,
bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True,
)
# turn into JSON
seq = processor.batch_decode(outputs.sequences)[0]
seq = seq.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
seq = re.sub(r"<.*?>", "", seq, count=1).strip() # remove first task start token
seq = processor.token2json(seq)
if has_metadata:
ground_truth = json.loads(sample["ground_truth"])
ground_truth = ground_truth["gt_parse"]
evaluator = JSONParseEvaluator()
score = evaluator.cal_acc(seq, ground_truth)
accs.append(score)
print(seq)
output_list.append(seq)
df = pd.DataFrame(map(lambda x: x.get('text_sequence', ''), output_list))
df.to_csv('out.tsv', sep='\t', header=False, index=False)
if has_metadata:
scores = {"accuracies": accs, "mean_accuracy": np.mean(accs)}
print(scores, f"length : {len(accs)}")
print("Mean accuracy:", np.mean(accs))
main(config)