trustmark
trustmark copied to clipboard
TrustMark - Universal Watermarking for Arbitrary Resolution Images
``` conda create --name trustmark python=3.10 conda activate trustmark conda install pytorch cudatoolkit=12.8 -c pytorch -c conda-forge pip install torch==2.1.2 torchvision==0.16.2 -f https://download.pytorch.org/whl/torch_stable.html pip install . ``` This ends with:...
While you mentioned in the paper and the repository that 'Q' means the model leans toward visual quality and 'B' means it leans toward robustness, I couldn't find where MODE='C'...
Hi, I want to reproduce the results of your paper and train and evaluate the models on my own datasets and setups. Could you please provide me with the training...
``` ~/test/trustmark/rust$ cargo bench Compiling trustmark v0.1.0 (/home/neal/test/trustmark/rust) error: linking with `cc` failed: exit status: 1 [super long error message] logits_processor.cc:(.text._ZNSt10_HashtableIiiSaIiENSt8__detail9_IdentityESt8equal_toIiESt4hashIiENS1_18_Mod_range_hashingENS1_20_Default_ranged_hashENS1_20_Prime_rehash_policyENS1_17_Hashtable_traitsILb0ELb1ELb1EEEE9_M_rehashEmRKm[_ZNSt10_HashtableIiiSaIiENSt8__detail9_IdentityESt8equal_toIiESt4hashIiENS1_18_Mod_range_hashingENS1_20_Default_ranged_hashENS1_20_Prime_rehash_policyENS1_17_Hashtable_traitsILb0ELb1ELb1EEEE9_M_rehashEmRKm]+0x107): undefined reference to `std::__throw_bad_array_new_length()' /usr/bin/ld: /home/neal/test/trustmark/rust/target/release/deps/libort_sys-67043d96470b7a97.rlib(uni_dir_attn_lstm.cc.o): in function `onnxruntime::contrib::rnn::detail::UniDirectionalAttnLstm::Compute(gsl::span...
I haven't saw any detail regarding watermarking in only center portion of the image with respect to the image aspect ratio, then why you are you limiting the watermarking area...