diff --git a/modules/image/text_to_image/ernie_vilg/README.md b/modules/image/text_to_image/ernie_vilg/README.md index 823d859b1f71fd5050de273c1c11c5f9c3323131..36b33d613a3e3da096c88d09a8977e67d2d47a6c 100755 --- a/modules/image/text_to_image/ernie_vilg/README.md +++ b/modules/image/text_to_image/ernie_vilg/README.md @@ -21,7 +21,7 @@ |数据集|-| |是否支持Fine-tuning|否| |模型大小|-| -|最新更新日期|2022-08-02| +|最新更新日期|2022-10-14| |数据指标|-| ### 应用效果展示 @@ -78,7 +78,8 @@ - ```python def generate_image( text_prompts:str, - style: Optional[str] = "油画", + style: Optional[str] = "探索无限", + resolution: Optional[str] = "1024*1024", topk: Optional[int] = 6, output_dir: Optional[str] = 'ernievilg_output') ``` @@ -88,7 +89,9 @@ - **参数** - text_prompts(str): 输入的语句，描述想要生成的图像的内容。 - - style(Optional[str]): 生成图像的风格，当前支持'油画','水彩','粉笔画','卡通','儿童画','蜡笔画','探索无限'。 + - style(Optional[str]): 生成图像的风格，当前支持 古风、油画、水彩、卡通、二次元、浮世绘、蒸汽波艺术、 + low poly、像素风格、概念艺术、未来主义、赛博朋克、写实风格、洛丽塔风格、巴洛克风格、超现实主义、探索无限。 + - resolution(Optional[str]): 生成图像的分辨率，当前支持 '1024\*1024', '1024\*1536', '1536\*1024'，默认为'1024\*1024'。 - topk(Optional[int]): 保存前多少张图，最多保存6张。 - output_dir(Optional[str]): 保存输出图像的目录，默认为"ernievilg_output"。 @@ -100,14 +103,6 @@ ## 四、 Prompt 指南 -(在 GitHub 阅读 README 的用户可以访问 www.youpromptme.cn 获得更好的阅读体验) - -(图片比较多，完全加载页面可能需要 3min) - -作者：佳祥 (LCL-Brew) - -原文地址： https://github.com/OleNet/YouPromptMe/tree/gh-pages/you-prompt-me - 这是一份如何调整 Prompt 得到更漂亮的图片的经验性文档。我们的结果和经验都来源于[文心 ERNIE-ViLG Demo](https://wenxin.baidu.com/moduleApi/ernieVilg) 和[社区的资料](#related-work)。 @@ -797,6 +792,9 @@ DiscoDiffusion Prompt 技巧资料：https://docs.google.com/document/d/1l8s7uS2 初始发布 +* 1.1.0 + + 增加分辨率参数以及所支持的风格 ```shell - $ hub install ernie_vilg == 1.0.0 + $ hub install ernie_vilg == 1.1.0 ``` diff --git a/modules/image/text_to_image/ernie_vilg/module.py b/modules/image/text_to_image/ernie_vilg/module.py index f812c60cce24d17372ed29a1a97efb3651c65ae4..f08ae13a2d9ad031ad6e4a3a65db83d134c6d8c4 100755 --- a/modules/image/text_to_image/ernie_vilg/module.py +++ b/modules/image/text_to_image/ernie_vilg/module.py @@ -21,7 +21,7 @@ from paddlehub.module.module import serving @moduleinfo(name="ernie_vilg", - version="1.0.0", + version="1.1.0", type="image/text_to_image", summary="", author="baidu-nlp", @@ -64,7 +64,8 @@ class ErnieVilG: def generate_image(self, text_prompts, - style: Optional[str] = "油画", + style: Optional[str] = "探索无限", + resolution: Optional[str] = "1024*1024", topk: Optional[int] = 6, visualization: Optional[bool] = True, output_dir: Optional[str] = 'ernievilg_output'): @@ -72,7 +73,9 @@ class ErnieVilG: Create image by text prompts using ErnieVilG model. :param text_prompts: Phrase, sentence, or string of words and phrases describing what the image should look like. - :param style: Image stype, currently supported 油画、水彩、粉笔画、卡通、儿童画、蜡笔画、探索无限。 + :param style: Image stype, currently supported 古风、油画、水彩、卡通、二次元、浮世绘、蒸汽波艺术、 + low poly、像素风格、概念艺术、未来主义、赛博朋克、写实风格、洛丽塔风格、巴洛克风格、超现实主义、探索无限。 + :param resolution: Resolution of images, currently supported "1024*1024", "1024*1536", "1536*1024". :param topk: Top k images to save. :param visualization: Whether to save images or not. :output_dir: Output directory @@ -91,7 +94,8 @@ class ErnieVilG: data={ 'access_token': token, "text": text_prompt, - "style": style + "style": style, + "resolution": resolution }) res = res.json() if res['code'] == 4001: @@ -113,7 +117,8 @@ class ErnieVilG: data={ 'access_token': token, "text": text_prompt, - "style": style + "style": style, + "resolution": resolution }) res = res.json() if res['code'] != 0: @@ -128,11 +133,31 @@ class ErnieVilG: start_time = time.time() process_bar = tqdm(total=100, unit='%') results = {} - first_iter = True + total_time = 60 * len(taskids) while True: + end_time = time.time() + duration = end_time - start_time + progress_rate = int((duration) / total_time * 100) + if not taskids: + progress_rate = 100 + if progress_rate > process_bar.n: + if progress_rate >= 100: + if not taskids: + increase_rate = 100 - process_bar.n + else: + increase_rate = 0 + else: + increase_rate = progress_rate - process_bar.n + else: + increase_rate = 0 + process_bar.update(increase_rate) + if duration < 30: + time.sleep(5) + continue + else: + time.sleep(6) if not taskids: break - total_time = 0 has_done = [] for taskid in taskids: res = requests.post(get_url, @@ -177,17 +202,6 @@ class ErnieVilG: else: print(res['msg']) raise RuntimeError(res['msg']) - total_time = int(re.match('[0-9]+', str(res['data']['waiting'])).group(0)) * 60 - end_time = time.time() - progress_rate = int(((end_time - start_time) / total_time * 100)) if total_time != 0 else 100 - if progress_rate > process_bar.n: - increase_rate = progress_rate - process_bar.n - if progress_rate >= 100: - increase_rate = 100 - process_bar.n - else: - increase_rate = 0 - process_bar.update(increase_rate) - time.sleep(5) for taskid in has_done: taskids.remove(taskid) print('Saving Images...') @@ -228,6 +242,7 @@ class ErnieVilG: self.token = self._apply_token(self.ak, self.sk) results = self.generate_image(text_prompts=args.text_prompts, style=args.style, + resolution=args.resolution, topk=args.topk, visualization=args.visualization, output_dir=args.output_dir) @@ -254,9 +269,17 @@ class ErnieVilG: self.arg_input_group.add_argument('--text_prompts', type=str) self.arg_input_group.add_argument('--style', type=str, - default='油画', - choices=['油画', '水彩', '粉笔画', '卡通', '儿童画', '蜡笔画', '探索无限'], + default='探索无限', + choices=[ + '古风', '油画', '水彩', '卡通', '二次元', '浮世绘', '蒸汽波艺术', 'low poly', '像素风格', '概念艺术', + '未来主义', '赛博朋克', '写实风格', '洛丽塔风格', '巴洛克风格', '超现实主义', '探索无限' + ], help="绘画风格") + self.arg_input_group.add_argument('--resolution', + type=str, + default='1024*1024', + choices=['1024*1024', '1024*1536', '1536*1024'], + help="图像分辨率") self.arg_input_group.add_argument('--topk', type=int, default=6, help="选取保存前多少张图，最多10张") self.arg_input_group.add_argument('--ak', type=str, default=None, help="申请文心api使用token的ak") self.arg_input_group.add_argument('--sk', type=str, default=None, help="申请文心api使用token的sk")