Faster-rcnn.pytorch: Cannot even make the demo work...help...

[isalirezag]

I am pretty disappointed in myself. Appologies for my stupid question in advance. I am pretty new to all these detection stuff and pytorch thing.

I went through the way that you explained to run the demo.
I guess i have everything set as you said.
I have a data folder, inside of it i have two folders (pretrained_model and VOCdevkit2007). the VOCdevkit2007 is created via softlink.
In the pretrained_model folder I have resnet101_caffe.pth and vgg16_caffe.pth.

I am trying to run the demo to run it on the images that you provided on the images folder.
So I am going to run this command:

python demo.py --net vgg16 \ --checksession $SESSION --checkepoch $EPOCH --checkpoint $CHECKPOINT \ --cuda --load_dir path/to/model/directoy

and i run it like this:

python demo.py --net vgg16 \ --checksession 1 --checkepoch 6 --checkpoint 416 \ --cuda --load_dir /data/pretrained_model/vgg16_caffe.pth

am i doing anything wrong so far?
This is what happened and it is not working:
Called with args: Namespace(batch_size=1, cfg_file='cfgs/vgg16.yml', checkepoch=6, checkpoint=416, checksession=1, class_agnostic=False, cuda=True, dataset='pascal_voc', image_dir='images', load_dir='/data/pretrained_model/vgg16_caffe.pth', mGPUs=False, net='vgg16', parallel_type=0, set_cfgs=None, vis=False, webcam_num=-1) Using config: {'ANCHOR_RATIOS': [0.5, 1, 2], 'ANCHOR_SCALES': [8, 16, 32], 'CROP_RESIZE_WITH_MAX_POOL': False, 'CUDA': False, 'DATA_DIR': '/home/alireza/Desktop/Code/VID/FRCNN2/faster-rcnn.pytorch/data', 'DEDUP_BOXES': 0.0625, 'EPS': 1e-14, 'EXP_DIR': 'vgg16', 'FEAT_STRIDE': [16], 'GPU_ID': 0, 'MATLAB': 'matlab', 'MAX_NUM_GT_BOXES': 20, 'MOBILENET': {'DEPTH_MULTIPLIER': 1.0, 'FIXED_LAYERS': 5, 'REGU_DEPTH': False, 'WEIGHT_DECAY': 4e-05}, 'PIXEL_MEANS': array([[[102.9801, 115.9465, 122.7717]]]), 'POOLING_MODE': 'align', 'POOLING_SIZE': 7, 'RESNET': {'FIXED_BLOCKS': 1, 'MAX_POOL': False}, 'RNG_SEED': 3, 'ROOT_DIR': '/home/alireza/Desktop/Code/VID/FRCNN2/faster-rcnn.pytorch', 'TEST': {'BBOX_REG': True, 'HAS_RPN': True, 'MAX_SIZE': 1000, 'MODE': 'nms', 'NMS': 0.3, 'PROPOSAL_METHOD': 'gt', 'RPN_MIN_SIZE': 16, 'RPN_NMS_THRESH': 0.7, 'RPN_POST_NMS_TOP_N': 300, 'RPN_PRE_NMS_TOP_N': 6000, 'RPN_TOP_N': 5000, 'SCALES': [600], 'SVM': False}, 'TRAIN': {'ASPECT_GROUPING': False, 'BATCH_SIZE': 256, 'BBOX_INSIDE_WEIGHTS': [1.0, 1.0, 1.0, 1.0], 'BBOX_NORMALIZE_MEANS': [0.0, 0.0, 0.0, 0.0], 'BBOX_NORMALIZE_STDS': [0.1, 0.1, 0.2, 0.2], 'BBOX_NORMALIZE_TARGETS': True, 'BBOX_NORMALIZE_TARGETS_PRECOMPUTED': True, 'BBOX_REG': True, 'BBOX_THRESH': 0.5, 'BG_THRESH_HI': 0.5, 'BG_THRESH_LO': 0.0, 'BIAS_DECAY': False, 'BN_TRAIN': False, 'DISPLAY': 10, 'DOUBLE_BIAS': True, 'FG_FRACTION': 0.25, 'FG_THRESH': 0.5, 'GAMMA': 0.1, 'HAS_RPN': True, 'IMS_PER_BATCH': 1, 'LEARNING_RATE': 0.01, 'MAX_SIZE': 1000, 'MOMENTUM': 0.9, 'PROPOSAL_METHOD': 'gt', 'RPN_BATCHSIZE': 256, 'RPN_BBOX_INSIDE_WEIGHTS': [1.0, 1.0, 1.0, 1.0], 'RPN_CLOBBER_POSITIVES': False, 'RPN_FG_FRACTION': 0.5, 'RPN_MIN_SIZE': 8, 'RPN_NEGATIVE_OVERLAP': 0.3, 'RPN_NMS_THRESH': 0.7, 'RPN_POSITIVE_OVERLAP': 0.7, 'RPN_POSITIVE_WEIGHT': -1.0, 'RPN_POST_NMS_TOP_N': 2000, 'RPN_PRE_NMS_TOP_N': 12000, 'SCALES': [600], 'SNAPSHOT_ITERS': 5000, 'SNAPSHOT_KEPT': 3, 'SNAPSHOT_PREFIX': 'res101_faster_rcnn', 'STEPSIZE': [30000], 'SUMMARY_INTERVAL': 180, 'TRIM_HEIGHT': 600, 'TRIM_WIDTH': 600, 'TRUNCATED': False, 'USE_ALL_GT': True, 'USE_FLIPPED': True, 'USE_GT': False, 'WEIGHT_DECAY': 0.0005}, 'USE_GPU_NMS': True} Traceback (most recent call last): File "demo.py", line 163, in <module> raise Exception('There is no input directory for loading network from ' + input_dir) Exception: There is no input directory for loading network from /data/pretrained_model/vgg16_caffe.pth/vgg16/pascal_voc

Can anyone please help me .... Thanks

[cbasavaraj]

Hi, you have to distinguish between a pre-trained model like vgg16 or res101 (which is the base for faster-rcnn) and the model you train, which will be used in demo.py. So first run a training, which will give you a model, and you can then use this model to run the demo.

[isalirezag]

@cbasavaraj I dont have a good GPU at the moment to train my network.
I thought if i download the pre trained model from here I should be able to run the demo. is not that possible?
Is there anyway to try the demo without training?

[cbasavaraj]

You can try Google Cloud, they give you USD 250 worth free compute time. Train your model there.
This repo doesn't have a trained faster-rcnn. Not sure about other places.

[isalirezag]

@cbasavaraj Thank you for your help. I will try google cloud.
but in the repository what they mean when they said:

Pretrained Model
We used two pretrained models in our experiments, VGG and ResNet101. You can download these two models from:
VGG16: Dropbox, VT Server
ResNet101: Dropbox, VT Server
Download them and put them into the data/pretrained_model/.

what are these pretrained model useful for?

[cbasavaraj]

These models are pre-trained on ImageNet images for classification only. So you use them as the backbone for faster-rcnn which is an object detection architecture that can use many different underlying CNNs. Faster-RCNN itself has to be trained by you now for object detection.

[Michael-Hsu]

@isalirezag I am also a freshman to PyTorch and Faster R-CNN. And I also encounter the same problem.I'm wondering if you have solved this problem!