train bug

[zpz915]

python train.py \

--dataset mpiigaze \ --snapshot output/snapshots \ --gpu 0 \ --num_epochs 50 \ --batch_size 16 \ --lr 0.00001 \ --arch ResNet101 \ --alpha 1 Loading data. 45000 items removed from dataset that have an angle > 0 Traceback (most recent call last): File "train.py", line 276, in train_loader_gaze = DataLoader( File "/root/miniconda3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 268, in init sampler = RandomSampler(dataset, generator=generator) File "/root/miniconda3/lib/python3.8/site-packages/torch/utils/data/sampler.py", line 102, in init raise ValueError("num_samples should be a positive integer " ValueError: num_samples should be a positive integer value, but got num_samples=0

[zpz915]

i read the code,find dataset=Mpiigaze(testlabelpathombined,args.gazeMpiimage_dir, transformations, True, fold) give 5 paramters ,but class Mpiigaze(Dataset): def init(self, pathorg, root, transform, train, angle,fold=0): accept 6 paramters ,is there any bug?

[zpz915]

when i guess angel=180,the train also meet other bug: Traceback (most recent call last): File "train.py", line 299, in {'params': get_ignored_params(model, args.arch), 'lr': 0}, TypeError: get_ignored_params() takes 1 positional argument but 2 were given

[yakhyo]

hi @zpz915 , have you succeeded training the model. I am also facing the same issue.

[forlayo]

Same issue here, any solutions ?

[yakhyo]

@forlayo , I succeeded training Gaze360 and going to release it soon and adding mobilenet new backbone as well.

[tiamo405]

I tried to download the dataset but it seems like the dataset is too big, can anyone split the dataset for me around 5gb so I can download it?

[yakhyo]

@tiamo405 , MPIIFaceGaze is relatively smaller than Gaze360.

[tiamo405]

@tiamo405 , MPIIFaceGaze is relatively smaller than Gaze360.

MPIIFaceGaze not folder Label, link download: https://www.mpi-inf.mpg.de/departments/computer-vision-and-machine-learning/research/gaze-based-human-computer-interaction/its-written-all-over-your-face-full-face-appearance-based-gaze-estimation

[tiamo405]

link folder 940Mb: http://datasets.d2.mpi-inf.mpg.de/MPIIGaze/MPIIFaceGaze.zip

[ipsampling]

I have identified and resolved issues in the code, and understand why using the Gaze360 dataset can result in successful training while using the MPIIFaceGaze dataset reports errors. This is because MPIIFaceGaze data processing lacks an angle parameter, and all data with angles less than 0 degrees are filtered out.