Spaces:
Running
Running
File size: 5,537 Bytes
0d735a2 febe0d2 0d735a2 eb2e91c 0d735a2 54c1afe 0d735a2 709bd90 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 |
import os
os.system('cd fairseq;'
'pip install ./; cd ..')
os.system('ls -l')
import torch
import numpy as np
import re
from fairseq import utils,tasks
from fairseq import checkpoint_utils
from fairseq import distributed_utils, options, tasks, utils
from fairseq.dataclass.utils import convert_namespace_to_omegaconf
from utils.zero_shot_utils import zero_shot_step
from tasks.mm_tasks.vqa_gen import VqaGenTask
from models.ofa import OFAModel
from PIL import Image
from torchvision import transforms
import gradio as gr
# Register VQA task
tasks.register_task('vqa_gen',VqaGenTask)
# turn on cuda if GPU is available
use_cuda = torch.cuda.is_available()
# use fp16 only when GPU is available
use_fp16 = False
os.system('wget https://ofa-silicon.oss-us-west-1.aliyuncs.com/checkpoints/ofa_large_384.pt; '
'mkdir -p checkpoints; mv ofa_large_384.pt checkpoints/ofa_large_384.pt')
# specify some options for evaluation
parser = options.get_generation_parser()
input_args = ["", "--task=vqa_gen", "--beam=100", "--unnormalized", "--path=checkpoints/ofa_large_384.pt", "--bpe-dir=utils/BPE"]
args = options.parse_args_and_arch(parser, input_args)
cfg = convert_namespace_to_omegaconf(args)
# Load pretrained ckpt & config
task = tasks.setup_task(cfg.task)
models, cfg = checkpoint_utils.load_model_ensemble(
utils.split_paths(cfg.common_eval.path),
task=task
)
# Move models to GPU
for model in models:
model.eval()
if use_fp16:
model.half()
if use_cuda and not cfg.distributed_training.pipeline_model_parallel:
model.cuda()
model.prepare_for_inference_(cfg)
# Initialize generator
generator = task.build_generator(models, cfg.generation)
# Image transform
from torchvision import transforms
mean = [0.5, 0.5, 0.5]
std = [0.5, 0.5, 0.5]
patch_resize_transform = transforms.Compose([
lambda image: image.convert("RGB"),
transforms.Resize((cfg.task.patch_image_size, cfg.task.patch_image_size), interpolation=Image.BICUBIC),
transforms.ToTensor(),
transforms.Normalize(mean=mean, std=std),
])
# Text preprocess
bos_item = torch.LongTensor([task.src_dict.bos()])
eos_item = torch.LongTensor([task.src_dict.eos()])
pad_idx = task.src_dict.pad()
# Normalize the question
def pre_question(question, max_ques_words):
question = question.lower().lstrip(",.!?*#:;~").replace('-', ' ').replace('/', ' ')
question = re.sub(
r"\s{2,}",
' ',
question,
)
question = question.rstrip('\n')
question = question.strip(' ')
# truncate question
question_words = question.split(' ')
if len(question_words) > max_ques_words:
question = ' '.join(question_words[:max_ques_words])
return question
def encode_text(text, length=None, append_bos=False, append_eos=False):
s = task.tgt_dict.encode_line(
line=task.bpe.encode(text),
add_if_not_exist=False,
append_eos=False
).long()
if length is not None:
s = s[:length]
if append_bos:
s = torch.cat([bos_item, s])
if append_eos:
s = torch.cat([s, eos_item])
return s
# Construct input for open-domain VQA task
def construct_sample(image: Image, question: str):
patch_image = patch_resize_transform(image).unsqueeze(0)
patch_mask = torch.tensor([True])
question = pre_question(question, task.cfg.max_src_length)
question = question + '?' if not question.endswith('?') else question
src_text = encode_text(' {}'.format(question), append_bos=True, append_eos=True).unsqueeze(0)
src_length = torch.LongTensor([s.ne(pad_idx).long().sum() for s in src_text])
ref_dict = np.array([{'yes': 1.0}]) # just placeholder
sample = {
"id":np.array(['42']),
"net_input": {
"src_tokens": src_text,
"src_lengths": src_length,
"patch_images": patch_image,
"patch_masks": patch_mask,
},
"ref_dict": ref_dict,
}
return sample
# Function to turn FP32 to FP16
def apply_half(t):
if t.dtype is torch.float32:
return t.to(dtype=torch.half)
return t
# Function for image captioning
def open_domain_vqa(Image, Question):
sample = construct_sample(Image, Question)
sample = utils.move_to_cuda(sample) if use_cuda else sample
sample = utils.apply_to_sample(apply_half, sample) if use_fp16 else sample
# Run eval step for open-domain VQA
with torch.no_grad():
result, scores = zero_shot_step(task, generator, models, sample)
return result[0]['answer']
title = "OFA-Visual_Question_Answering"
description = "Gradio Demo for OFA-Visual_Question_Answering. Upload your own image (high-resolution images are recommended) or click any one of the examples, and click " \
"\"Submit\" and then wait for OFA's answer. "
article = "<p style='text-align: center'><a href='https://github.com/OFA-Sys/OFA' target='_blank'>OFA Github " \
"Repo</a></p> "
examples = [['cat-4894153_1920.jpg', 'where are the cats?'], ['men-6245003_1920.jpg', 'how many people are in the image?'], ['labrador-retriever-7004193_1920.jpg', 'what breed is the dog in the picture?'], ['Starry_Night.jpeg', 'what style does the picture belong to?']]
io = gr.Interface(fn=open_domain_vqa, inputs=[gr.inputs.Image(type='pil'), "textbox"], outputs=gr.outputs.Textbox(label="Answer"),
title=title, description=description, article=article, examples=examples,
allow_flagging=False, allow_screenshot=False)
io.launch(share=True) |