train-image-classifier docs assumes wrong structure for classes

[dgtlmoon]

Looking at https://www.deepdetect.com/server/docs/train-image-classifier/ in #751 I discovered that its expecting to look in subdirectories, but the ZIP structure from deepdetect with the example is just

dog.0.jpg
cat.1.jpg

etc etc etc

But I'm pretty sure deepdetect is expecting

dog/0.jpg
cat/1.jpg

Where "data":["/opt/models/cats_dogs/images/"] or to be more specific /opt/models/cats_dogs/images/dog/0.jpg

Looking at https://www.deepdetect.com/dd/datasets/cats_dogs.zip

Archive:  cats_dogs.zip
  Length      Date    Time    Name
---------  ---------- -----   ----
        0  2013-09-20 10:12   train/
    12414  2013-09-20 10:05   train/cat.0.jpg
    16880  2013-09-20 10:05   train/cat.1.jpg
    34315  2013-09-20 10:05   train/cat.10.jpg
    28377  2013-09-20 10:05   train/cat.100.jpg
     5944  2013-09-20 10:05   train/cat.1000.jpg
    21944  2013-09-20 10:05   train/cat.10000.jpg
    27322  2013-09-20 10:05   train/cat.10001.jpg
    25723  2013-09-20 10:05   train/cat.10002.jpg
    28035  2013-09-20 10:05   train/cat.10003.jpg
    12973  2013-09-20 10:05   train/cat.10004.jpg
     8245  2013-09-20 10:05   train/cat.10005.jpg

I discovered this because the stats on training were not improving and the only class as a prediction was images... which is really not any of my class names but it was in the path in data

[dgtlmoon]

I can confirm this works (sub-directory name per class) and that https://www.deepdetect.com/dd/datasets/cats_dogs.zip has it all in one directory