facebookresearch/dino

Issues

• How to get the feature map by using pretrained model.

  #239 opened 2 years ago by JiuqingDong
  2

• Help!Why the results of 'attention_visual.py' are different from those in the paper？

  #228 opened 2 years ago by Rorschach-xyz
  1

• Loss is not decresing

  #273 opened 10 months ago by risingClouds
  5

• Linear Evaluation LR Sweep Values

  #249 opened a year ago by ferreirafabio
  2

• Changing number of classes

  #282 opened 5 months ago by vikashg
  0

• Reproducing results on segmentation using training from scratch

  #281 opened 5 months ago by sonamsingh19
  0

• A question about DINOHead

  #244 opened 2 years ago by wenhaoli-xmu
  2

• How to convert to onnx

  #276 opened 8 months ago by Ochre-amber
  2

• Why Teacher network perform bettert than Student one during training?

  #274 opened 9 months ago by Hongbo-Z
  1

• Choice of out_dim

  #279 opened 6 months ago by JunboShen
  1

• Can I use dino model to match images with pixels? Just like with the clip model you can match pixels with text.

  #278 opened 7 months ago by dajiebi
  1

• Cannot reproduce KNN performance for vanilla ViT-S training

  #280 opened 6 months ago by wangh09
  1

• torch hubconf.py is missing a function

  #243 opened 2 years ago by nkondapa
  3

• How small batch sizes affect performance

  #224 opened 2 years ago by bryanwong17
  1

• How to concatenate attention maps from different heads

  #237 opened 2 years ago by LLL-YUE
  1

• Best features for image similarity

  #272 opened 10 months ago by fbliman
  1

• Faster training

  #247 opened 2 years ago by rbareja25
  2

• Question regarding patch retrieval demo

  #250 opened 8 months ago by crypdick
  2

• problem when trying to finetune a model, using resnet50 backbone

  #251 opened a year ago by saeidrazavi
  1

• Loss doesn't decrease

  #246 opened 2 years ago by AwayFromKeyboardwarrior
  6

• Why do we skip cases where the student and teacher operate on the same view? If they are operating on different views, why should they produce similar results to calculate the cross-entropy loss?

  #267 opened a year ago by jinghere11
  1

• Image resolution at intermediate layers?

  #252 opened a year ago by yousafe007
  3

• a solution to solve memory issues (but slows down training a bit)

  #269 opened a year ago by fawazsammani
  0

• Why not setting correct img_size when building the student network?

  #266 opened a year ago by LichunZhang
  1

• For large batches (256), there is a problem of loss non convergence

  #265 opened a year ago by Elijah-Yi
  0

• Difference between DINO and DINOv2

  #264 opened a year ago by Tulsishah
  0

• Loss Nan Error

  #263 opened a year ago by Elijah-Yi
  0

• Why the patchembedding defaut img size is not equal to the image size in visualize attention?

  #261 opened a year ago by LWShowTime
  1

• sock.connect(sa)TimeoutError : raise err[Errno 110] Connection timed out

  #233 opened 2 years ago by wdc233
  1

• Facing Accuracy Drop while running in QDQ onnx runtime

  #260 opened a year ago by akarshit15092000
  0

• Intermediate checkpoints?

  #259 opened a year ago by MLAlex1
  0

• issue with the ViT-S/8 full checkpoint

  #256 opened a year ago by melikaqh
  0

• Hyperparameters for ViT-B/16 + ImageNet pretraining

  #255 opened a year ago by kalyani7195
  0

• The averaged learning rate in log.txt

  #254 opened a year ago by haohang96
  0

• starting from checkpoint

  #253 opened a year ago by hyunjung00
  0

• about the DINO training loss

  #225 opened 2 years ago by Faisal-Hajari
  1

• Number of classes

  #223 opened 2 years ago by VGrondin
  1

• dinohead weight normalized

  #248 opened a year ago by kffeng
  0

• How to use teacher weights?

  #245 opened 2 years ago by zahrasa
  0

• RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)`

  #242 opened 2 years ago by PapaMadeleine2022
  0

• when extracting features from image, i see that the program use the teacher network, can I use the student network, if using the sutdent network, weather the performance will be reduced?What's the difference between the feature of student network and the feature of teacher network？

  #241 opened 2 years ago by yyy809643573
  0

• how to train DINO with MSE loss.

  #240 opened 2 years ago by Backdrop9019
  0

• Could you tell me how to visualize PCA 3-channel image just like the showcase in your README

  #236 opened 2 years ago by shoutOutYangJie
  0

• How does a zero KLD indicate a collapse?

  #235 opened 2 years ago by YoojLee
  0

• normalize knn probabilities

  #234 opened 2 years ago by clemsgrs
  0

• Memory Occupancy

  #232 opened 2 years ago by develop-productivity
  0

• Training on uncurated/unbound datasets

  #231 opened 2 years ago by ajkailash
  0

• Can model from torch.hub.load support differnet img_size from training?

  #230 opened 2 years ago by Luciennnnnnn
  0

• Some questions about dino

  #229 opened 2 years ago by mdatres
  0

• Supervised Fine-Tuning of Teacher / Student Transformer Weights

  #226 opened 2 years ago by MatCorr
  0