Datasets

[Uaena-cf]

I'm very interested in this dataset. Could you please share the download link or provide instructions on how to access it?

[Britton-Li]

I'm very interested in this dataset. Could you please share the download link or provide instructions on how to access it?

Same request!!!

[WangYuXuan2022]

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly.
As for the instructions, our dataset primarily consists of three parts:
1.public datasets:
a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing)
b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html)
2.web-crawled images from the Internet
3.synthetic images by latest model(e.g, Flux-dev)
In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

[Uaena-cf]

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly. As for the instructions, our dataset primarily consists of three parts: 1.public datasets: a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing) b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html) 2.web-crawled images from the Internet 3.synthetic images by latest model(e.g, Flux-dev) In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

Thank you so much for your response!!

I hope you don't mind me asking one more thing—do you still happen to have the model weight files saved? If you do, would it be possible to share them? If not, no worries at all, and I apologize for the trouble.

Once again, thank you so much—your work is truly amazing!!

[Britton-Li]

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly. As for the instructions, our dataset primarily consists of three parts: 1.public datasets: a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing) b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html) 2.web-crawled images from the Internet 3.synthetic images by latest model(e.g, Flux-dev) In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

Another question: Can you provide some instructions for CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0) dataset. I check this huggingface link, but find unaccessible. Great thanks!!!

[WangYuXuan2022]

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly. As for the instructions, our dataset primarily consists of three parts: 1.public datasets: a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing) b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html) 2.web-crawled images from the Internet 3.synthetic images by latest model(e.g, Flux-dev) In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

Thank you so much for your response!!

I hope you don't mind me asking one more thing—do you still happen to have the model weight files saved? If you do, would it be possible to share them? If not, no worries at all, and I apologize for the trouble.

Once again, thank you so much—your work is truly amazing!!

Sorry again, both dataset and model weights are stored on the company's server, and I currently do not have access to download or retrieve them locally. Since our code is mainly based on Diffusers, you may download backbone model (e.g, sdxl-base, flux-dev) and implement on your dataset.

[WangYuXuan2022]

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly. As for the instructions, our dataset primarily consists of three parts: 1.public datasets: a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing) b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html) 2.web-crawled images from the Internet 3.synthetic images by latest model(e.g, Flux-dev) In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

Another question: Can you provide some instructions for CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0) dataset. I check this huggingface link, but find unaccessible. Great thanks!!!

I've checked the huggingface link, it looks like you need to request and get access through email. I directly use the dataset that had already been downloaded by my colleagues at the company before.

[Uaena-cf]

Thank you very much for your answer. I really appreciate it!

I sincerely apologize that this dataset was constructed during my internship as part of the company's resources. Since I have already left the company, I'm unable to provide the link directly. As for the instructions, our dataset primarily consists of three parts: 1.public datasets: a. single-person images: CosmicManHQ-1.0(https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0), Human-centric Dataset(https://drive.google.com/drive/folders/1jWOouxul6LZWHsU1NsH2RvZemzfgR3PN?usp=sharing) b. multi-person images: BodyHands(http://vision.cs.stonybrook.edu/~supreeth/BodyHands/BodyHands.zip), CrowdHuman(https://www.crowdhuman.org/download.html) 2.web-crawled images from the Internet 3.synthetic images by latest model(e.g, Flux-dev) In addition, we apply VLM(e.g, InterVL, QWen) and yolo detector(face & hands) for captioning and annotating.

Thank you so much for your response!!
I hope you don't mind me asking one more thing—do you still happen to have the model weight files saved? If you do, would it be possible to share them? If not, no worries at all, and I apologize for the trouble.
Once again, thank you so much—your work is truly amazing!!

Sorry again, both dataset and model weights are stored on the company's server, and I currently do not have access to download or retrieve them locally. Since our code is mainly based on Diffusers, you may download backbone model (e.g, sdxl-base, flux-dev) and implement on your dataset.

[Uaena-cf]

I'm very interested in this dataset. Could you please share the download link or provide instructions on how to access it?

Same request!!!

Hey man, are you working on reproducing this paper too? Maybe we could exchange some ideas!

[Britton-Li]

I'm very interested in this dataset. Could you please share the download link or provide instructions on how to access it?

Same request!!!

Hey man, are you working on reproducing this paper too? Maybe we could exchange some ideas!

Of course! I’m currently working on reproducing this paper. However, I’ve found that the most challenging part is obtaining the dataset. As far as I know, the source dataset used in this paper — CosmicManHQ-1.0 — is now inaccessible, which makes reproduction quite difficult. Please feel free to reach out via the email listed in my profile if you’d like to discuss this further.

[Uaena-cf]

I'm very interested in this dataset. Could you please share the download link or provide instructions on how to access it?

Same request!!!

Hey man, are you working on reproducing this paper too? Maybe we could exchange some ideas!

Of course! I’m currently working on reproducing this paper. However, I’ve found that the most challenging part is obtaining the dataset. As far as I know, the source dataset used in this paper — CosmicManHQ-1.0 — is now inaccessible, which makes reproduction quite difficult. Please feel free to reach out via the email listed in my profile if you’d like to discuss this further.

Hello! Have you been training this method recently? When I train with my own dataset, I notice that step_loss keeps oscillating and doesn’t seem to be decreasing— is this normal? Also, around step 15 000 the gradients become NaN and the training stops. Have you ever encountered this? How did you solve it? Thank you very much!