Didn't get step 3 in Five step pipeline?

[prashant334]

@kPsarakis @achilleasvlogiaris @arkajitb @psoilis

"Extract SVM patches: same as step 1 but with different arguments."

[kPsarakis]

Hi @prashant334

this was a mistake on our part, let me clear out the confusion. The "new" steps are the following (we will also change them in the README):

1. Extract CNN training patches: as shown in extract_patches.py

2. Train CNN: open the train_net.py and change DATA_DIR to point to the patches path extracted from the previous step. Run the script, it will save the trained network as shown in

   Image-Forgery-Detection-CNN/src/train_net.py

   Line 30 in 9dc3f04

   torch.save(cnn.state_dict(), 'Cnn.pt')

3. Compute image features: as shown in feature_extraction.py. Here you will need to provide the trained CNN as input from the previous step. Change the path in the following line

   Image-Forgery-Detection-CNN/src/feature_extraction.py

   Line 8 in 9dc3f04

   model.load_state_dict(torch.load('../data/output/pre_trained_cnn/CASIA2_WithRot_LR001_b128_nodrop.pt',

4. Run SVM cross-validation: change the features path in svm_classification.py at line 5 to point to the latest feature extraction. Run the script.

the different steps are also described in the report. Section 3.3 refers to the first two steps, section 3.4 to the rest. The hyperparameters that we used and the experiment pipeline are described in section 4.

apologies again for the confusion, we hope it is more clear now.

[prashant334]

Okay.... I Intuitively implemented in this way only.Thanks for validation. On Jul 28, 2020 00:31, "Kyriakos Psarakis" <notifications@github.com> wrote: Hi @prashant334 <https://github.com/prashant334> this was a mistake on our part, let me clear out the confusion. The "new" steps are the following (we will also change them in the README): 1. Extract CNN training patches: as shown in extract_patches.py 2. Train CNN: open the train_net.py and change DATA_DIR to point to the patches path extracted from the previous step. Run the script, it will save the trained network as shown in https://github.com/kPsarakis/ Image-Forgery-Detection-CNN/blob/9dc3f0468cf5f0161f2ac296ece980 01a603c1ea/src/train_net.py#L30 <https://github.com/kPsarakis/Image-Forgery-Detection-CNN/blob/9dc3f0468cf5f0161f2ac296ece98001a603c1ea/src/train_net.py#L30> 3. Compute image features: as shown in feature_extraction.py. Here you will need to provide the trained CNN as input from the previous step. Change the path in the following line https://github.com/kPsarakis/ Image-Forgery-Detection-CNN/blob/9dc3f0468cf5f0161f2ac296ece980 01a603c1ea/src/feature_extraction.py#L8 <https://github.com/kPsarakis/Image-Forgery-Detection-CNN/blob/9dc3f0468cf5f0161f2ac296ece98001a603c1ea/src/feature_extraction.py#L8> 4. Run SVM cross-validation: change the features path in svm_classification.py at line 5 to point to the latest feature extraction. Run the script. the different steps are also described in the report <https://github.com/kPsarakis/Image-Forgery-Detection-CNN/blob/master/reports/Group_10-Image_Forgery_Detection_report.pdf>. Section 3.3 refers to the first two steps, section 3.4 to the rest. The hyperparameters that we used and the experiment pipeline are described in section 4. apologies again for the confusion, we hope it is more clear now. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE7HLG6EIKX3AQKSZ637NH3R5XFHXANCNFSM4PDX4OLQ> .