Training and validation loss curve remains flat when trained on MVTec data

[vishwanathvenkat]

Greetings,
I trained the network on MVTEC tile data (Tried on various other datasets as well). The details are as follows

Command

python <Root folder>/train.py -d <data folder>/capsule -a mvtecCAE -l ssim -b 32

config.py

ROT_ANGLE = 5
W_SHIFT_RANGE = 0.05
H_SHIFT_RANGE = 0.05
FILL_MODE = "nearest"
BRIGHTNESS_RANGE = [0.95, 1.05]
VAL_SPLIT = 0.2

# Learning Rate Finder parameters
START_LR = 1e-7
LR_MAX_EPOCHS = 10
LRF_DECREASE_FACTOR = 0.85

# Training parameters
EARLY_STOPPING = 12
REDUCE_ON_PLATEAU = 6

# Finetuning parameters
FINETUNE_SPLIT = 0.1
STEP_MIN_AREA = 5
START_MIN_AREA = 5
STOP_MIN_AREA = 1005

Env

appdirs==1.4.4
argon2-cffi==20.1.0
astunparse==1.6.3
async-generator==1.10
attrs==20.3.0
backcall==0.2.0
black==19.10b0
bleach==3.2.0
CacheControl==0.12.6
cachetools==4.2.1
cchardet==2.1.7
certifi==2020.12.5
cffi==1.14.5
chardet==4.0.0
click==7.1.2
colorama==0.4.3
contextlib2==0.6.0
cycler==0.10.0
decorator==4.4.2
distlib==0.3.0
distro==1.4.0
fastprogress==1.0.0
filelock==3.0.12
flatbuffers==1.12
fvcore==0.1.3.post20210226
gast==0.3.3
google-auth==1.27.0
google-auth-oauthlib==0.4.2
google-pasta==0.2.0
grpcio==1.32.0
h5py==2.10.0
html5lib==1.0.1
idna==2.10
imageio==2.9.0
iopath==0.1.4
ipaddr==2.2.0
ipython==7.21.0
ipython-genutils==0.2.0
jedi==0.18.0
jieba==0.42.1
joblib==1.0.1
Keras==2.4.3
keras-bert==0.86.0
keras-embed-sim==0.8.0
keras-layer-normalization==0.14.0
keras-multi-head==0.27.0
keras-pos-embd==0.11.0
keras-position-wise-feed-forward==0.6.0
Keras-Preprocessing==1.1.2
keras-self-attention==0.46.0
keras-transformer==0.38.0
kiwisolver==1.3.1
ktrain==0.25.4
langdetect==1.0.8
lockfile==0.12.2
Markdown==3.3.4
matplotlib==3.3.4
msgpack==0.6.2
networkx==2.5
numpy==1.20.1
oauthlib==3.1.0
opt-einsum==3.3.0
packaging==20.9
pandas==1.2.2
parso==0.8.1
pathspec==0.8.1
pep517==0.8.2
pexpect==4.8.0
pickleshare==0.7.5
Pillow==8.1.1
pkg-resources==0.0.0
portalocker==2.2.1
progress==1.5
prompt-toolkit==3.0.16
protobuf==3.15.3
ptyprocess==0.7.0
pyasn1==0.4.8
pyasn1-modules==0.2.8
pycparser==2.20
Pygments==2.8.0
pyparsing==2.4.7
python-dateutil==2.8.1
pytoml==0.1.21
pytz==2021.1
PyWavelets==1.1.1
PyYAML==5.4.1
regex==2020.11.13
requests==2.25.1
requests-oauthlib==1.3.0
retrying==1.3.3
rsa==4.7.2
sacremoses==0.0.43
scikit-image==0.18.1
scikit-learn==0.23.2
scipy==1.6.1
sentencepiece==0.1.91
seqeval==0.0.19
six==1.15.0
SSIM-PIL==1.0.12
syntok==1.3.1
tabulate==0.8.9
tensorboard==2.4.1
tensorboard-plugin-wit==1.8.0
tensorflow-estimator==2.4.0
tensorflow-gpu==2.4.1
termcolor==1.1.0
threadpoolctl==2.1.0
tifffile==2021.2.26
tokenizers==0.9.3
toml==0.10.2
tqdm==4.58.0
traitlets==5.0.5
transformers==3.5.1
typed-ast==1.4.2
typing-extensions==3.7.4.3
urllib3==1.26.3
wcwidth==0.2.5
webencodings==0.5.1
Werkzeug==1.0.1
Whoosh==2.7.4
wrapt==1.12.1
yacs==0.1.8

Machine details

Ubuntu 20.04.2 LTS
Memory: 15.5GB
GPU: NVIDIA Corporation GM107M [GeForce GTX 960M]

Issue

No significant learning is happening

Sample output

Additional info

I am getting similar results on all other cases in Mvtec dataset too.
Having other questions like why is val loss less than training loss, etc.

I am getting a feeling that I am not setting something right.
Kindly help.

[shahardror]

I can confirm, experiencing the same results with the carpet data set.
Any guidance would be appreciated

[sanersbug]

@vizav @shahardror Do you solved it , i have the same error

[j2233ack]

@vizav @shahardror
Do you solved it??
If so, how to solve it
THX