Eval7 - MSCOCO Poisoning Dataset

[yusong-tan]

For object detection poisoning scenario

[swsuggs]

@lcadalzo If you could just check that this works for you, that would be great

[swsuggs]

Going to change the name of the dataset as discussed in the poisoning group meeting

[lcadalzo]

I can confirm that I can load the dataset and retrieve a batch of data successfully:

>>> from armory.data.datasets import mscoco_poisoning                                                                                   
2023-04-05 20:47:46 48s INFO     armory.delayed_imports:<module>:7 Importing and configuring torch, tensorflow, and art, if available. T
his may take some time.                                                                                                                 
^[[B^[[B^[[B^[[B^[[B^[[B^[[B^[[B2023-04-05 20:47:52 54s INFO     art.config:set_data_path:55 set ART_DATA_PATH to /tmp/.art/data        
2023-04-05 20:47:54 56s INFO     art.config:set_data_path:55 set ART_DATA_PATH to /armory/saved_models/art                              
>>> ds = mscoco_poisoning()                                                                                                             
2023-04-05 20:48:06 1m8s INFO     armory.data.utils:download_verify_dataset_cache:251 Attempting download_verigy_dataset_cache with data
set_dir                                                                                                                                 
2023-04-05 20:48:06 1m8s INFO     armory.data.utils:download_verify_dataset_cache:282 Downloading dataset: mscoco_poisoning...          
/opt/conda/lib/python3.9/site-packages/urllib3/connectionpool.py:1043: InsecureRequestWarning: Unverified HTTPS request is being made to
 host 'armory-public-data.s3.us-east-2.amazonaws.com'. Adding certificate verification is strongly advised. See: https://urllib3.readthe
docs.io/en/1.26.x/advanced-usage.html#ssl-warnings                                                                                      
  warnings.warn(                                                                                                                        
2023-04-05 20:49:41 2m43s INFO     armory.data.utils:download_requests:195 downloaded /armory/datasets/cache/mscoco_poisoning_1.0.0.tar.
gz from https://armory-public-data.s3.us-east-2.amazonaws.com/mscoco_poisoning/mscoco_poisoning_1.0.0.tar.gz                            
2023-04-05 20:49:41 2m43s INFO     armory.data.utils:download_verify_dataset_cache:299 Verifying sha256 hash of download...             
2023-04-05 20:49:46 2m48s INFO     armory.data.utils:download_verify_dataset_cache:315 Extracting .tfrecord files from download...      
mscoco_poisoning/                                                                                                                       
mscoco_poisoning/2017/                                                                                                                  
mscoco_poisoning/2017/1.0.0/                                                                                                            
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00004-of-00016                                                              
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00014-of-00016                                                              
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00013-of-00016                                                              
mscoco_poisoning/2017/1.0.0/objects-label.labels.txt                                                                                    
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00003-of-00016                                                              
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00005-of-00016                                                              
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00015-of-00016                                                              
mscoco_poisoning/2017/1.0.0/features.json                                                                                               
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00012-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00002-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00011-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00001-of-00016
mscoco_poisoning/2017/1.0.0/dataset_info.json
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00006-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00009-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00010-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00000-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00008-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-train.tfrecord-00007-of-00016
mscoco_poisoning/2017/1.0.0/mscoco_poisoning-validation.tfrecord-00000-of-00001
mscoco_poisoning/.config/
mscoco_poisoning/.config/metadata.json
>>> x, y = next(ds)
>>> type(x)
<class 'numpy.ndarray'>
>>> x.shape
(1, 238, 640, 3)
>>> type(y)
<class 'list'>
>>> len(y)
1
>>> type(y[0])
<class 'dict'>
>>> y[0].keys()
dict_keys(['area', 'id', 'is_crowd', 'boxes', 'labels'])
>>> y[0]['boxes'].shape
(1, 4)
>>> y[0]['labels'].shape                                             
(1,)