Datasets:
Is it possible to provide the original image dataset?
This dataset is comprehensive and valuable for various computer vision tasks related to anime images, such as image compression and edge detection. However, the current dataset includes salt and pepper noise and has a relatively low resolution. Therefore, I would like to inquire if it is possible to provide the original image dataset?
I have uploaded the dataset to https://huggingface.co/datasets/nyuuzyou/AnimeHeadsv3/tree/main/AHv3-NA. Additionally, I have included more version information in the README and updated the dataset upload script. However, I'm experiencing an issue where the images are not displaying correctly on Hugging Face. I will investigate this matter further.
Hi, thank you very much for your sharing. I think that I found the dataset your shared on Roboflow.
I used some of the original images your provided in I created Anime Character Object Detection dataset, combined with the AniDet3 dataset, to achieve impressive object detection results. The online demo can be found here: https://huggingface.co/spaces/deepghs/anime_object_detection, and the PyPI library is available here: https://deepghs.github.io/imgutils/main/api_doc/detect/person.html.
In addition to that, I am also annotating a dataset for detecting anime character heads to further improve the existing model. Is it possible to collaborate on Roboflow? I can upload some data and assist with annotation using the uploaded existing model.
Hello, I apologize for the long response. I recently finished AnimeHeadsv3 training, which I thought might be useful for annotating your AniDet3 dataset. Interestingly, the anime images in AnimeHeadsv3 were annotated with the v2 model, and it performed quite well. This suggests that the v3 model, which can be downloaded from https://huggingface.co/nyuuzyou/AnimeHeads, might be useful for creating a new dataset and further improving the accuracy and speed of the annotation process.
Although I am not interested in collaborating at the moment, I encourage you to explore the potential of the AnimeHeadsv3 model for your project. I hope it will prove to be a valuable resource for your work.
I wish you the best of luck with your annotation process and future projects.
nterestingly, the anime images in AnimeHeadsv3 were annotated with the v2 model, and it performed quite well.
Exactly, for target detection tasks, it is a very efficient and effective way to use old models to assist new model annotations, and Roboflow provides this feature.
Here is our annotated dataset: https://huggingface.co/datasets/deepghs/anime_face_detection
I wish it can help. We have already deploy the model trained with combined dataset (merging our data set and your data set) into use. We will continue to collect new pictures for training, and will continue to improve the data in the future Set disease improves model quality. :smile: