LutingWang / OADP

Object-Aware Distillation Pyramid for Open-Vocabulary Object Detection
Apache License 2.0
54 stars 3 forks source link

阿里网盘中为什么没有文件? #4

Closed Originlightwkp closed 1 year ago

LutingWang commented 1 year ago

阿里云盘不支持分享二进制文件,我们正在把数据迁移到百度网盘和谷歌云盘上

ZJUJeffLai commented 1 year ago

It seems I can only find globals.tar.gz and blocks.tar.gz in the shared folder but cannot find objects.tar.gz. Also, I found I cannot tar them. Could you help check if the right files are uploaded to the shared folder?

LutingWang commented 1 year ago

It seems I can only find globals.tar.gz and blocks.tar.gz in the shared folder but cannot find objects.tar.gz. Also, I found I cannot tar them. Could you help check if the right files are uploaded to the shared folder?

We have double-checked that both globals.tar.gz and blocks.tar.gz can be decompressed via tar -zxvf ${FILENAME}. For your reference, the MD5 checksum for globals.tar.gz is 6c9c9a7517c667e8320fb2e55292cd3a, and for blocks.tar.gz is cd5cfb5e0dd9279d73d32ea6cf491e12.

The object features (>100GB) are much larger than the globals.tar.gz and blocks.tar.gz. Therefore, we recommend regenerating these features by yourself, which may be faster. We are trying to upload the features, but may take some time to finish.

ZJUJeffLai commented 1 year ago

Thanks! After we generate object features, do we still need to run these?

cd objects/train2017 ls > objects split -d -3000 - objects. < objects for i in objects.[0-9][0-9]; do zip -q -9 "$i.zip" -@ < "$i" mv "$i.zip" ../.. done rm objects*

LutingWang commented 1 year ago

Thanks! After we generate object features, do we still need to run these?

cd objects/train2017 ls > objects split -d -3000 - objects. < objects for i in objects.[0-9][0-9]; do zip -q -9 "$i.zip" -@ < "$i" mv "$i.zip" ../.. done rm objects*

You are welcome. No, the commands are used for compressing the object features, before uploading to Baidu Netdisk. If you are generating object features, you can completely ignore these commands. Otherwise, if you will be using our pre-computed features, the commands may provide useful information on how to decompress the tar files.

DaqiLin commented 1 year ago

OADP/data ├── coco │ └── proposals │ ├── rpn_r101_fpn_coco_train.pkl │ ├── rpn_r101_fpn_coco_val.pkl │ ├── oln_r50_fpn_coco_train.pkl │ └── oln_r50_fpn_coco_val.pkl └── lvis_v1 └── proposals ├── oln_r50_fpn_lvis_train.pkl └── oln_r50_fpn_lvis_val.pkl I can't find "oln_r50_fpn_lvis_train.pkl" in the Baidu Netdisk

LutingWang commented 1 year ago

OADP/data ├── coco │ └── proposals │ ├── rpn_r101_fpn_coco_train.pkl │ ├── rpn_r101_fpn_coco_val.pkl │ ├── oln_r50_fpn_coco_train.pkl │ └── oln_r50_fpn_coco_val.pkl └── lvis_v1 └── proposals ├── oln_r50_fpn_lvis_train.pkl └── oln_r50_fpn_lvis_val.pkl I can't find "oln_r50_fpn_lvis_train.pkl" in the Baidu Netdisk

Uploaded. But lvis support is still under testing. We will release the config and checkpoint for lvis training soon.