17c81454创建于 3月10日历史提交

文件	最后提交记录	最后更新时间
README.md	新增DeOldify模型 Co-authored-by: socrahow1<suzihao4@h-partners.com> # message auto-generated for no-merge-commit merge: !7498 merge master into master 新增DeOldify模型 Created-by: socrahow1 Commit-by: socrahow1 Merged-by: ascend-robot Description: ## Motivation Adaptation DeOldify model ## Modification Adaptation DeOldify model ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist Before PR: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized After PR: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/ModelZoo-PyTorch!7498	2 个月前
diff.patch	新增DeOldify模型 Co-authored-by: socrahow1<suzihao4@h-partners.com> # message auto-generated for no-merge-commit merge: !7498 merge master into master 新增DeOldify模型 Created-by: socrahow1 Commit-by: socrahow1 Merged-by: ascend-robot Description: ## Motivation Adaptation DeOldify model ## Modification Adaptation DeOldify model ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist Before PR: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized After PR: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/ModelZoo-PyTorch!7498	2 个月前
infer.py	新增DeOldify模型 Co-authored-by: socrahow1<suzihao4@h-partners.com> # message auto-generated for no-merge-commit merge: !7498 merge master into master 新增DeOldify模型 Created-by: socrahow1 Commit-by: socrahow1 Merged-by: ascend-robot Description: ## Motivation Adaptation DeOldify model ## Modification Adaptation DeOldify model ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist Before PR: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized After PR: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/ModelZoo-PyTorch!7498	2 个月前

DeOldify(aclgraph)-推理指导

DeOldify(aclgraph)-推理指导
概述
推理环境准备
快速上手

概述

DeOldify 是一种著名的自动上色开源算法。该模型采用 ResNet 作为编码器构建了具有 UNet 结构的网络，并提出了几个不同的训练版本，在效果、效率、鲁棒性等方面具有良好的综合性能。

推理环境准备

配套	版本	环境指导准备
固件与驱动	25.5.1	Pytorch框架推理环境准备
CANN	8.5.0	-
torch	2.9.0	-
torch-npu	2.9.0	-
torchvision	0.24.0	-
说明：该模型暂不适用于Atlas 300I系列产品	-	-

快速上手

环境配置

安装依赖

git clone https://gitcode.com/ascend/ModelZoo-PyTorch.git
cd ModelZoo-PyTorch/ACL_PyTorch/built-in/cv/DeOldify
git clone https://github.com/jantic/DeOldify.git(或者git clone https://gitcode.com/gh_mirrors/de/DeOldify.git)
cd DeOldify
patch -p1 < ../diff.patch
apt-get update
apt-get install -y libgl1 libglib2.0-0 ffmpeg
pip install -r requirements.txt
cd ..
cp infer.py DeOldify
cd DeOldify
pip uninstall triton-ascend

获取权重

下载DeOldify模型权重，放置于本地目录models中

artistic
stable
video

获取测试数据

下载测试图片，放置于本地目录test_images中

cd test_images
wget https://media.githubusercontent.com/media/dana-kelley/DeOldify/refs/heads/master/test_images/1860Girls.jpg
cd ..

如需做视频上色，则下载测试视频，放置于本地目录video/source中（如果没有就自行创建目录）

模型推理

设置环境变量

# 开启虚拟内存
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
# 队列优化特性
export TASK_QUEUE_ENABLE=1
# CPU绑核
export CPU_AFFINITY_CONF=1

推理命令

ASCEND_RT_VISIBLE_DEVICES=X python infer.py \
--image_artistic \
--image \
--video \
--image_path "test_images/XXX.jpg" \
--image_url "https://XXX.jpg" \
--video_path "XX.mp4" \
--video_url "httpsL//XX.mp4" \
--image_render_factor 35 \
--video_render_factor 21 \
--watermarked \
--warm_num 2 \
--infer_num 3

参数说明
ASCEND_RT_VISIBLE_DEVICES：推理时使用的卡，仅支持单卡，仅能输入0，1，2这样的数字。
image_artistic：图片上色时是否使用artistic模型。
image：进行图片上色。
video：进行视频上色。
image_path：本地带上色图片路径，与image_url同时使用时不生效。
image_url：带上色图片网址，优先级高于image_path，两个参数共存时image_path不生效。
video_path：本地带上色视频路径，与video_url同时使用时不生效。
video_url：带上色图片网址，优先级高于video_path，两个参数共存时video_path不生效。
image_render_factor：图片渲染因子，数值越高，着色品质越好，但是处理时间越长，默认值35。
video_render_factor：视频渲染因子，数值越高，着色品质越好，但是处理时间越长，默认值21。
watermarked：添加水印。
warm_num：预热次数，预热可提升模型推理性能。
infer_num：每个输入推理次数。

比如推理图片

ASCEND_RT_VISIBLE_DEVICES=0 python infer.py \
--image_artistic \
--image \
--image_path "test_images/1860Girls.jpg" \
--image_render_factor 35

第一次推理时会自动下载所需的辅助模型模型下载.png

结果展示

本次测试使用Atlas 800I A2 64G单卡，64核CPU，在quay.io/ascend/vllm-ascend:v0.15.0rc1上推理结果如下：

推理性能

图片尺寸	封装格式	render_factor	并发数	推理时长
720p	png	35	1	0.6s
1080p	jpg	35	1	0.3s

视频规格	封装格式	render_factor	并发数	推理时长
20s 720p@30fps	mp4	21	1	105s