tonyngjichun / solar Goto Github PK

Hi,
do you support vgg16 pretrained test?
if yes how can I run it?
what *.pth should I use?

parser.add_argument('--network', '-n', metavar='NETWORK', default='resnet101-solar-best.pth',
help="network to be evaluated. " )

alternatively,
can you explain how to train with vgg16?
I tried setting --arch 'vgg16'
but it fails in networks.py in init() in this line:
last_feat_in = base_model.inplanes
orch.nn.modules.module.ModuleAttributeError: 'VGG' object has no attribute 'inplanes'

Thanks

When I extract 1-million distractors descriptors, I find that the revisitop1m.txt file contains more images than the number of images extracted from 100 tar.gz files ( I download the http://ptak.felk.cvut.cz/revisitop/revisitop1m/), so I get an error when running the program（python3 -m solar_global.examples.extract_1m）. I would like to ask how you deal with this problem. Thank you for your reply.

How do I prepare/structure a custom dataset?

Is there any chance to release the exact steps used to train the local solar system on UBC?

I am reading you paper but i can not understand the difference between SOS and FOS.
By the way ,can you release you training code?
thank you!

Hi, @tonyngjichun

When training with the GL18 dataset, what is the whitening dataset?

Did you ever use "retrieval-SfM-120k-whiten.pkl" from the sfm-120k dataset?
When whitening, it's not clear, but in your code it looks like you're using "retrieval-SfM-120k-whiten.pkl", not the GL18 dataset. right? (

thanks

Hi~
Are there any experimental results using the sfm-120k train set?
There are signs in your code that you want to use sfm-120k as the train set.
(https://github.com/tonyngjichun/SOLAR/blob/master/solar_global/examples/train.py#L40)

In my purely personal opinion, I'm noticing that many studies in this field have different train sets and same evaluation sets. I don't think this is fair.
I think both should be evaluated comparatively in the same case.
So, I'd like to know for a fair comparison.
By any chance, do you have any experimental results?
thanks~
@tonyngjichun

how can i visualize feature map like you . thanks

Hi, I have selected gl18-tl-resnet50-gem-w, how can I convert the vector dimension to 512?

By default, gl18-tl-resnet50-gem-w outputs a vector dimension of 2048, but what if I want to convert it to 512 dimensions? What should I do?

I cant access to the https://imperialcollegelondon.box.com
can you provide the model from the other website

hi, when I install model with: wget -nc https://imperialcollegelondon.box.com/shared/static/fznpeayct6btel2og2wjjgvqw0ziqnk4.pth -O data/networks/resnet101-solar-best.pth

,
It seems that the link timed out。
Is there any other way to download the model?

thanks

Did the author train on resnet-50 backbone? If did, could you tell me the results on ROxford and RParis? Thanks you very much!

Hi, thank you for sharing this project. Good job! I tried to run this project, but I met some questions.

1. When I run python3 create_db_pickle.py, the logs are as follow:
   
   Is the number of training images and val images correct?
   2 I meet a problem when training:
   
   Have you ever had a similar problem? I try to decrease batch size, but it doesn't work for me.
   Can you tell me what is the specific version of pytorch, cuda and cudnn you use?

HI, i am facing a problem, need your help .
when i load the resnet101-solar-best.pth on pytorch 1.6+cu101, is OK. i switch to pytorch 1.10+cuda11.3 , then problem
occurs 'RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory' . Do you know why?

I use SOLAR for vehicle re-identification/pedestrian re-identification. but generate custom dataset is wrong.
After the network converges, the negative sample l2 distance is close to 0 in create_epoch_tuples.
In the evaluation, rank1 is close to 100%, but mAP is very low.

# create db pickle
_,_,image_paths,file_ids,labels = gil('custom_train.csv','/home/lxk/ZHP/data/VeIDData/VERI',True)

for mode in ['train', 'val']:
    image_list = train_idx_list if mode == 'train' else val_idx_list
  
    # boxes_dict = boxes[mode]

    for i,list in tqdm(image_list.items()):
        for idx in list:
            positives = []

            db_dict[mode]['cids'].append(image_paths[idx]) # image path

            db_dict[mode]['cluster'].append(labels[idx])   # class label

            pidxs_potential = [i for i in list]

            try:
                pidxs_potential.remove(idx)
            except:
                pass

            if len(pidxs_potential) == 0:
                continue

            pidxs = np.random.choice(pidxs_potential, min(len(pidxs_potential), 1)).tolist()

            db_dict[mode]['bbxs'].append(None) # bbox none

            db_dict[mode]['qidxs'].append(idx)    #anchor image idx
            db_dict[mode]['pidxs'].append(pidxs[0])  #postive image idx


save_path = './db_gl18.pkl'

pickle.dump(db_dict, open(save_path, 'wb'))

# in class TuplesBatchedDataset(data.Dataset): def __init__
self.images = [os.path.join(self.ims_root, db['cids'][i]+'.jpg') for i in range(len(db['cids']))]
#modified to 
self.images = [db['cids'][i] for i in range(len(db['cids']))]