113 lines
4.7 KiB
Python
113 lines
4.7 KiB
Python
#
|
|
# Copyright 2019 The FATE Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
import torch
|
|
from transformers import \
|
|
TableTransformerForObjectDetection,\
|
|
AutoImageProcessor
|
|
from PIL import ImageDraw
|
|
from random import randint
|
|
|
|
|
|
class TableTransformer:
|
|
def __init__(self,
|
|
rec_mdlnm="microsoft/table-transformer-structure-recognition"):
|
|
"""
|
|
If you have trouble downloading HuggingFace models, -_^ this might help!!
|
|
|
|
For Linux:
|
|
export HF_ENDPOINT=https://hf-mirror.com
|
|
|
|
For Windows:
|
|
Good luck
|
|
^_-
|
|
|
|
"""
|
|
self.rec_img_pro = AutoImageProcessor.from_pretrained(rec_mdlnm)
|
|
self.rec_mdl = TableTransformerForObjectDetection.from_pretrained(
|
|
rec_mdlnm)
|
|
|
|
if torch.cuda.is_available():
|
|
self.rec_mdl.cuda()
|
|
self.batch_size = 1 # batch_size
|
|
|
|
def __friendly(self, batch_res, id2label):
|
|
res = []
|
|
for r in batch_res:
|
|
feas = []
|
|
for score, label, box in zip(r["scores"], r["labels"], r["boxes"]):
|
|
if label.item() == 0:
|
|
continue
|
|
box = [round(x, 2) for x in box.tolist()]
|
|
feas.append({
|
|
"type": id2label[label.item()],
|
|
"score": score.item(),
|
|
"bbox": box
|
|
})
|
|
res.append(feas)
|
|
return res
|
|
|
|
def __draw(self, bres, imgs, id2label):
|
|
for i, (img, r) in enumerate(zip(imgs, bres)):
|
|
draw = ImageDraw.Draw(img, "RGB")
|
|
for score, label, box in zip(r["scores"], r["labels"], r["boxes"]):
|
|
if label.item() == 0:
|
|
continue
|
|
r = randint(0, 255)
|
|
g = randint(0, 255)
|
|
b = randint(0, 255)
|
|
x0, y0, x1, y1 = box[0], box[1], box[2], box[-1]
|
|
draw.rectangle((x0, y0, x1, y1), outline=(r, g, b), width=1)
|
|
draw.text((x0, y0), id2label[label.item(
|
|
)] + ":{:.2f}".format(score), fill=(r, g, b))
|
|
img.save(f"./t{i}.%d.jpg" % randint(0, 1000))
|
|
|
|
def __call__(self, images, threshold=0.8):
|
|
res = []
|
|
for i in range(0, len(images), self.batch_size):
|
|
imgs = images[i: i + self.batch_size]
|
|
inputs = self.rec_img_pro(imgs, return_tensors="pt")
|
|
inputs = {k: inputs[k].to(self.rec_mdl.device)
|
|
if isinstance(inputs[k], torch.Tensor)
|
|
else inputs[k] for k in inputs.keys()}
|
|
outputs = self.rec_mdl(**inputs)
|
|
target_sizes = torch.tensor([img.size[::-1] for img in imgs])
|
|
# [scores, labels, boxes}]
|
|
with torch.no_grad():
|
|
bres = self.rec_img_pro.post_process_object_detection(outputs,
|
|
threshold=threshold,
|
|
target_sizes=target_sizes)
|
|
#self.__draw(bres, imgs, self.rec_mdl.config.id2label)
|
|
res.extend(self.__friendly(bres, self.rec_mdl.config.id2label))
|
|
return res
|
|
|
|
def detect(self, images):
|
|
res = []
|
|
for i in range(0, len(images), self.batch_size):
|
|
imgs = images[i: i + self.batch_size]
|
|
inputs = self.det_img_pro(imgs, return_tensors="pt")
|
|
inputs = {k: inputs[k].to(self.det_mdl.device)
|
|
if isinstance(inputs[k], torch.Tensor)
|
|
else inputs[k] for k in inputs.keys()}
|
|
outputs = self.det_mdl(**inputs)
|
|
target_sizes = torch.tensor([img.size[::-1] for img in imgs])
|
|
# [scores, labels, boxes}]
|
|
with torch.no_grad():
|
|
res.extend(self.__friendly(self.det_img_pro.post_process_object_detection(outputs,
|
|
threshold=0.9,
|
|
target_sizes=target_sizes),
|
|
self.det_mdl.config.id2label
|
|
))
|
|
return res
|