Wrong Checkpoint in Model Zoo (Multimodal)

OpenGVLab / unmasked_teacher

[ICCV2023 Oral] Unmasked Teacher: Towards Training-Efficient Video Foundation Models

https://arxiv.org/abs/2303.16058

MIT License

267 stars 13 forks source link

Wrong Checkpoint in Model Zoo (Multimodal) #26

Closed HyeongminLEE closed 7 months ago

HyeongminLEE commented 7 months ago

Hi, I'm trying to download and analyze the checkpoints you opened. During analyzing, I found that a wrong ckpt has been uploaded.

In the b16_25m.pth in https://github.com/OpenGVLab/unmasked_teacher/blob/main/multi_modality/MODEL_ZOO.md, There's no keys vision_proj, text_proj, itm_head, clip_teacher and it includes text_decoder which other checkpoints do not include. Is it intended? Or not, it would be helpful for improving this repo to re-uploading the ckpt.

Thanks!

Andy1621 commented 7 months ago

Have you downloaded the correct models? It seemed normal when I checked.

HyeongminLEE commented 7 months ago

Following the link on the table shared at OpenGVLab/unmasked_teacher/blob/main/multi_modality/MODEL_ZOO.md, I downloaded the b16_25m and b16_17m models by following command.

wget https://pjlab-gvm-data.oss-cn-shanghai.aliyuncs.com/umt/multi_modality/b16_25m.pth wget https://pjlab-gvm-data.oss-cn-shanghai.aliyuncs.com/umt/multi_modality/b16_17m.pth

Then I checked the keys of the two checkpoints. Then I got following result

Andy1621 commented 7 months ago

Thanks for your responses! We have updated the checkpoints. The previous checkpoints were overwritten by mistake.

HyeongminLEE commented 7 months ago

Thanks! It works now.