first sync

Author: Confusezius

example_benchmark_runs/example_benchmark_runs.sh

@@ -0,0 +1,40 @@
+#### Compute Default Results for ViT-B/32 and ViT-L/14.
+## For larger models, simply set e.g. --clip ViT-bigG-14
+
+# Compute Results on CIRR.
+# Note: This will create a submission JSON-file to upload to the evaluation server: https://cirr.cecs.anu.edu.au/test_process/
+# Note: To compute results on the validation split, simply set --split val
+datapath=/mnt/datasets_r/CIRR
+python src/main.py --dataset cirr --split test  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.contextual_modifier_prompt --clip ViT-B-32
+python src/main.py --dataset cirr --split test  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.contextual_modifier_prompt --clip ViT-L-14
+
+# Compute Results on the CIRCO test set.
+# Note: This will create a submission JSON-file to upload to the evaluation server: https://circo.micc.unifi.it/evaluation
+# Note: To compute results on the validation split, simply set --split val
+datapath=/mnt/datasets_r/CIRCO
+python src/main.py --dataset circo --split test  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt --clip ViT-B-32
+python src/main.py --dataset circo --split test  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt --clip ViT-L-14
+
+# Compute Results on each FASHIONIQ dataset. 
+# Note: FashionIQ results are reported on the validation set.
+datapath=/mnt/datasets_r/FASHIONIQ
+python src/main.py --dataset fashioniq_dress --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-B-32
+python src/main.py --dataset fashioniq_dress --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-L-14
+
+python src/main.py --dataset fashioniq_shirt --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-B-32
+python src/main.py --dataset fashioniq_shirt --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-L-14
+
+python src/main.py --dataset fashioniq_toptee --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-B-32
+python src/main.py --dataset fashioniq_toptee --split val  --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.structural_modifier_prompt_fashion --clip ViT-L-14
+
+### Compute Results on each GeneCIS benchmark
+# Note: GeneCIS results are reported on the validation set.
+datapath=/mnt/datasets_r/GENECIS
+# Change Attribute
+python src/main.py --dataset genecis_change_attribute --split val --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.short_modifier_prompt --clip ViT-B/32
+# Focus Attribute
+python src/main.py --dataset genecis_focus_attribute --split val --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.short_modifier_prompt --clip ViT-B/32
+# Change Object
+python src/main.py --dataset genecis_change_object --split val --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.short_modifier_prompt --clip ViT-B/32
+# Focus Object
+python src/main.py --dataset genecis_focus_object --split val --dataset-path $datapath --preload img_features captions mods --llm_prompt prompts.short_focus_object_modifier_prompt --clip ViT-B/32