COCO download script is broken
In the resulting folder,
-
annotations/instances_trainval35k.json.zipis corrupted (also why does the script downloads this but doesn't unzip it?) -
coco_labels.txtis missing (it can be found inssd.pytorch/data)
There are no errors in the console output.
You can download that file on here. https://dl.dropboxusercontent.com/s/s3tw5zcg7395368/instances_valminusminival2014.json.zip After download, just change file's name correctly(instances_trainval35k.json) and copy that to correct directory(annotations/).
Then, you can train with MS COCO dataset.
I think we can only train this model on coco but we can not test and evaluate, am I correctly understood Thank you
@seongkyun I have followed your answer on my MBP2022. I got this error. OSX 13.4.1 (c) (22F770820d) Python 3.11.3 My packages pip freeze.txt
/Users/sarit/data/coco
loading annotations into memory...
Traceback (most recent call last):
File "/Users/sarit/study/ssd.pytorch/train.py", line 256, in <module>
train()
File "/Users/sarit/study/ssd.pytorch/train.py", line 81, in train
dataset = COCODetection(root=args.dataset_root,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/sarit/study/ssd.pytorch/data/coco.py", line 91, in __init__
self.coco = COCO(osp.join(root, ANNOTATIONS,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/sarit/anaconda3/envs/try_openai/lib/python3.11/site-packages/pycocotools/coco.py", line 82, in __init__
dataset = json.load(f)
^^^^^^^^^^^^
File "/Users/sarit/anaconda3/envs/try_openai/lib/python3.11/json/__init__.py", line 293, in load
return loads(fp.read(),
^^^^^^^^^
File "<frozen codecs>", line 322, in decode
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xdc in position 69: invalid continuation byte
You can download that file on here. https://dl.dropboxusercontent.com/s/s3tw5zcg7395368/instances_valminusminival2014.json.zip After download, just change file's name correctly(instances_trainval35k.json) and copy that to correct directory(annotations/).
Then, you can train with MS COCO dataset.
You forgot to mention extract and rename it 👎
You can download that file on here. https://dl.dropboxusercontent.com/s/s3tw5zcg7395368/instances_valminusminival2014.json.zip After download, just change file's name correctly(instances_trainval35k.json) and copy that to correct directory(annotations/).
Then, you can train with MS COCO dataset.
Sorry but this link can't be found now. Can you please update it? Thanks!