Spaces:
Running
Running
File size: 3,934 Bytes
bc059ff |
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 |
import os
import torch
import torch.nn as nn
import numpy as np
import torch.nn.functional as F
import torchvision.transforms as T
from PIL import Image
from decord import VideoReader
from decord import cpu
from uniformer_light_video import uniformer_xxs_video
from kinetics_class_index import kinetics_classnames
from transforms import (
GroupNormalize, GroupScale, GroupCenterCrop,
Stack, ToTorchFormatTensor
)
import gradio as gr
from huggingface_hub import hf_hub_download
# Device on which to run the model
# Set to cuda to load on GPU
device = "cpu"
model_path = hf_hub_download(repo_id="Andy1621/uniformer_light", filename="uniformer_xxs16_160_k400.pth")
# Pick a pretrained model
model = uniformer_xxs_video()
state_dict = torch.load(model_path, map_location='cpu')
model.load_state_dict(state_dict)
# Set to eval mode and move to desired device
model = model.to(device)
model = model.eval()
# Create an id to label name mapping
kinetics_id_to_classname = {}
for k, v in kinetics_classnames.items():
kinetics_id_to_classname[k] = v
def get_index(num_frames, num_segments=8):
seg_size = float(num_frames - 1) / num_segments
start = int(seg_size / 2)
offsets = np.array([
start + int(np.round(seg_size * idx)) for idx in range(num_segments)
])
return offsets
def load_video(video_path):
vr = VideoReader(video_path, ctx=cpu(0))
num_frames = len(vr)
frame_indices = get_index(num_frames, 8)
# transform
crop_size = 160
scale_size = 160
input_mean = [0.485, 0.456, 0.406]
input_std = [0.229, 0.224, 0.225]
transform = T.Compose([
GroupScale(int(scale_size)),
GroupCenterCrop(crop_size),
Stack(),
ToTorchFormatTensor(),
GroupNormalize(input_mean, input_std)
])
images_group = list()
for frame_index in frame_indices:
img = Image.fromarray(vr[frame_index].asnumpy())
images_group.append(img)
torch_imgs = transform(images_group)
return torch_imgs
def inference(video):
vid = load_video(video)
# The model expects inputs of shape: B x C x H x W
TC, H, W = vid.shape
inputs = vid.reshape(1, TC//3, 3, H, W).permute(0, 2, 1, 3, 4)
with torch.no_grad():
prediction = model(inputs)
prediction = F.softmax(prediction, dim=1).flatten()
return {kinetics_id_to_classname[str(i)]: float(prediction[i]) for i in range(400)}
def set_example_video(example: list) -> dict:
return gr.Video.update(value=example[0])
demo = gr.Blocks()
with demo:
gr.Markdown(
"""
# UniFormer Light
Gradio demo for <a href='https://github.com/Sense-X/UniFormer' target='_blank'>UniFormer</a>: To use it, simply upload your video, or click one of the examples to load them. Read more at the links below.
"""
)
with gr.Box():
with gr.Row():
with gr.Column():
with gr.Row():
input_video = gr.Video(label='Input Video')
with gr.Row():
submit_button = gr.Button('Submit')
with gr.Column():
label = gr.Label(num_top_classes=5)
with gr.Row():
example_videos = gr.Dataset(components=[input_video], samples=[['./videos/hitting_baseball.mp4'], ['./videos/hoverboarding.mp4'], ['./videos/yoga.mp4']])
gr.Markdown(
"""
<p style='text-align: center'><a href='https://arxiv.org/abs/2201.09450' target='_blank'>[TPAMI] UniFormer: Unifying Convolution and Self-attention for Visual Recognition</a> | <a href='https://github.com/Sense-X/UniFormer' target='_blank'>Github Repo</a></p>
"""
)
submit_button.click(fn=inference, inputs=input_video, outputs=label)
example_videos.click(fn=set_example_video, inputs=example_videos, outputs=example_videos.components)
demo.launch(enable_queue=True) |