Spaces:
Running
Running
update font
Browse files- app.py +14 -14
- generate_plot.py +5 -5
app.py
CHANGED
|
@@ -20,7 +20,7 @@ i2t_models = [ # Average time spent running the following example
|
|
| 20 |
"gpt-4o-2024-05-13",
|
| 21 |
"llava-hf/llava-v1.6-vicuna-7b-hf"
|
| 22 |
]
|
| 23 |
-
perspectives = ["
|
| 24 |
main_scores_t2i = {}
|
| 25 |
main_scores_i2t = {}
|
| 26 |
|
|
@@ -32,22 +32,22 @@ for model in t2i_models:
|
|
| 32 |
for perspective in perspectives:
|
| 33 |
if perspective not in sub_scores_t2i.keys():
|
| 34 |
sub_scores_t2i[perspective] = {}
|
| 35 |
-
if perspective == "
|
| 36 |
main_scores_t2i[model][perspective] = hallucination_t2i_agg(model, "./data/results")["score"]
|
| 37 |
sub_scores_t2i[perspective][model] = hallucination_t2i_agg(model, "./data/results")["subscenarios"]
|
| 38 |
-
elif perspective == "
|
| 39 |
main_scores_t2i[model][perspective] = safety_t2i_agg(model, "./data/results")["score"]
|
| 40 |
sub_scores_t2i[perspective][model] = safety_t2i_agg(model, "./data/results")["subscenarios"]
|
| 41 |
-
elif perspective == "
|
| 42 |
main_scores_t2i[model][perspective] = adversarial_robustness_t2i_agg(model, "./data/results")["score"]
|
| 43 |
sub_scores_t2i[perspective][model] = adversarial_robustness_t2i_agg(model, "./data/results")["subscenarios"]
|
| 44 |
-
elif perspective == "
|
| 45 |
main_scores_t2i[model][perspective] = fairness_t2i_agg(model, "./data/results")["score"]
|
| 46 |
sub_scores_t2i[perspective][model] = fairness_t2i_agg(model, "./data/results")["subscenarios"]
|
| 47 |
-
elif perspective == "
|
| 48 |
main_scores_t2i[model][perspective] = privacy_t2i_agg(model, "./data/results")["score"]
|
| 49 |
sub_scores_t2i[perspective][model] = privacy_t2i_agg(model, "./data/results")["subscenarios"]
|
| 50 |
-
elif perspective == "
|
| 51 |
main_scores_t2i[model][perspective] = ood_t2i_agg(model, "./data/results")["score"]
|
| 52 |
sub_scores_t2i[perspective][model] = ood_t2i_agg(model, "./data/results")["subscenarios"]
|
| 53 |
else:
|
|
@@ -60,24 +60,24 @@ for model in i2t_models:
|
|
| 60 |
for perspective in perspectives:
|
| 61 |
if perspective not in sub_scores_i2t.keys():
|
| 62 |
sub_scores_i2t[perspective] = {}
|
| 63 |
-
if perspective == "
|
| 64 |
main_scores_i2t[model][perspective] = hallucination_i2t_agg(model, "./data/results")["score"]
|
| 65 |
sub_scores_i2t[perspective][model] = hallucination_i2t_agg(model, "./data/results")["subscenarios"]
|
| 66 |
-
elif perspective == "
|
| 67 |
main_scores_i2t[model][perspective] = safety_i2t_agg(model, "./data/results")["score"]
|
| 68 |
sub_scores_i2t[perspective][model] = safety_i2t_agg(model, "./data/results")["subscenarios"]
|
| 69 |
-
elif perspective == "
|
| 70 |
main_scores_i2t[model][perspective] = adversarial_robustness_i2t_agg(model, "./data/results")["score"]
|
| 71 |
sub_scores_i2t[perspective][model] = adversarial_robustness_i2t_agg(model, "./data/results")["subscenarios"]
|
| 72 |
-
elif perspective == "
|
| 73 |
main_scores_i2t[model][perspective] = fairness_i2t_agg(model, "./data/results")["score"]
|
| 74 |
sub_scores_i2t[perspective][model] = fairness_i2t_agg(model, "./data/results")["subscenarios"]
|
| 75 |
-
elif perspective == "
|
| 76 |
main_scores_i2t[model][perspective] = privacy_i2t_agg(model, "./data/results")["score"]
|
| 77 |
sub_scores_i2t[perspective][model] = privacy_i2t_agg(model, "./data/results")["subscenarios"]
|
| 78 |
-
elif perspective == "
|
| 79 |
main_scores_i2t[model][perspective] = ood_i2t_agg(model, "./data/results")["score"]
|
| 80 |
-
sub_scores_i2t[perspective][model] = ood_i2t_agg
|
| 81 |
else:
|
| 82 |
raise ValueError("Invalid perspective")
|
| 83 |
|
|
|
|
| 20 |
"gpt-4o-2024-05-13",
|
| 21 |
"llava-hf/llava-v1.6-vicuna-7b-hf"
|
| 22 |
]
|
| 23 |
+
perspectives = ["Safety", "Fairness", "Hallucination", "Privacy", "Adv", "OOD"]
|
| 24 |
main_scores_t2i = {}
|
| 25 |
main_scores_i2t = {}
|
| 26 |
|
|
|
|
| 32 |
for perspective in perspectives:
|
| 33 |
if perspective not in sub_scores_t2i.keys():
|
| 34 |
sub_scores_t2i[perspective] = {}
|
| 35 |
+
if perspective == "Hallucination":
|
| 36 |
main_scores_t2i[model][perspective] = hallucination_t2i_agg(model, "./data/results")["score"]
|
| 37 |
sub_scores_t2i[perspective][model] = hallucination_t2i_agg(model, "./data/results")["subscenarios"]
|
| 38 |
+
elif perspective == "Safety":
|
| 39 |
main_scores_t2i[model][perspective] = safety_t2i_agg(model, "./data/results")["score"]
|
| 40 |
sub_scores_t2i[perspective][model] = safety_t2i_agg(model, "./data/results")["subscenarios"]
|
| 41 |
+
elif perspective == "Adv":
|
| 42 |
main_scores_t2i[model][perspective] = adversarial_robustness_t2i_agg(model, "./data/results")["score"]
|
| 43 |
sub_scores_t2i[perspective][model] = adversarial_robustness_t2i_agg(model, "./data/results")["subscenarios"]
|
| 44 |
+
elif perspective == "Fairness":
|
| 45 |
main_scores_t2i[model][perspective] = fairness_t2i_agg(model, "./data/results")["score"]
|
| 46 |
sub_scores_t2i[perspective][model] = fairness_t2i_agg(model, "./data/results")["subscenarios"]
|
| 47 |
+
elif perspective == "Privacy":
|
| 48 |
main_scores_t2i[model][perspective] = privacy_t2i_agg(model, "./data/results")["score"]
|
| 49 |
sub_scores_t2i[perspective][model] = privacy_t2i_agg(model, "./data/results")["subscenarios"]
|
| 50 |
+
elif perspective == "OOD":
|
| 51 |
main_scores_t2i[model][perspective] = ood_t2i_agg(model, "./data/results")["score"]
|
| 52 |
sub_scores_t2i[perspective][model] = ood_t2i_agg(model, "./data/results")["subscenarios"]
|
| 53 |
else:
|
|
|
|
| 60 |
for perspective in perspectives:
|
| 61 |
if perspective not in sub_scores_i2t.keys():
|
| 62 |
sub_scores_i2t[perspective] = {}
|
| 63 |
+
if perspective == "Hallucination":
|
| 64 |
main_scores_i2t[model][perspective] = hallucination_i2t_agg(model, "./data/results")["score"]
|
| 65 |
sub_scores_i2t[perspective][model] = hallucination_i2t_agg(model, "./data/results")["subscenarios"]
|
| 66 |
+
elif perspective == "Safety":
|
| 67 |
main_scores_i2t[model][perspective] = safety_i2t_agg(model, "./data/results")["score"]
|
| 68 |
sub_scores_i2t[perspective][model] = safety_i2t_agg(model, "./data/results")["subscenarios"]
|
| 69 |
+
elif perspective == "Adv":
|
| 70 |
main_scores_i2t[model][perspective] = adversarial_robustness_i2t_agg(model, "./data/results")["score"]
|
| 71 |
sub_scores_i2t[perspective][model] = adversarial_robustness_i2t_agg(model, "./data/results")["subscenarios"]
|
| 72 |
+
elif perspective == "Fairness":
|
| 73 |
main_scores_i2t[model][perspective] = fairness_i2t_agg(model, "./data/results")["score"]
|
| 74 |
sub_scores_i2t[perspective][model] = fairness_i2t_agg(model, "./data/results")["subscenarios"]
|
| 75 |
+
elif perspective == "Privacy":
|
| 76 |
main_scores_i2t[model][perspective] = privacy_i2t_agg(model, "./data/results")["score"]
|
| 77 |
sub_scores_i2t[perspective][model] = privacy_i2t_agg(model, "./data/results")["subscenarios"]
|
| 78 |
+
elif perspective == "OOD":
|
| 79 |
main_scores_i2t[model][perspective] = ood_i2t_agg(model, "./data/results")["score"]
|
| 80 |
+
sub_scores_i2t[perspective][model] = ood_i2t_agg(model, "./data/results")["subscenarios"]
|
| 81 |
else:
|
| 82 |
raise ValueError("Invalid perspective")
|
| 83 |
|
generate_plot.py
CHANGED
|
@@ -52,11 +52,11 @@ def radar_plot(results, thetas, selected_models):
|
|
| 52 |
|
| 53 |
fig.add_trace(
|
| 54 |
go.Table(
|
| 55 |
-
header=dict(values=header_texts, font=dict(size=
|
| 56 |
cells=dict(
|
| 57 |
values=rows,
|
| 58 |
align="left",
|
| 59 |
-
font=dict(size=
|
| 60 |
height=30
|
| 61 |
),
|
| 62 |
# columnwidth=column_widths
|
|
@@ -120,11 +120,11 @@ def main_radar_plot(main_scores, selected_models):
|
|
| 120 |
|
| 121 |
fig.add_trace(
|
| 122 |
go.Table(
|
| 123 |
-
header=dict(values=header_texts, font=dict(size=
|
| 124 |
cells=dict(
|
| 125 |
values=rows,
|
| 126 |
align="left",
|
| 127 |
-
font=dict(size=
|
| 128 |
height=30,
|
| 129 |
),
|
| 130 |
columnwidth=column_widths,
|
|
@@ -206,7 +206,7 @@ if __name__ == "__main__":
|
|
| 206 |
"gpt-4o-2024-05-13",
|
| 207 |
"llava-hf/llava-v1.6-vicuna-7b-hf"
|
| 208 |
]
|
| 209 |
-
perspectives = ["
|
| 210 |
main_scores_t2i = {}
|
| 211 |
main_scores_i2t = {}
|
| 212 |
sub_scores_t2i = {}
|
|
|
|
| 52 |
|
| 53 |
fig.add_trace(
|
| 54 |
go.Table(
|
| 55 |
+
header=dict(values=header_texts, font=dict(size=14.5), align="left"),
|
| 56 |
cells=dict(
|
| 57 |
values=rows,
|
| 58 |
align="left",
|
| 59 |
+
font=dict(size=14.5),
|
| 60 |
height=30
|
| 61 |
),
|
| 62 |
# columnwidth=column_widths
|
|
|
|
| 120 |
|
| 121 |
fig.add_trace(
|
| 122 |
go.Table(
|
| 123 |
+
header=dict(values=header_texts, font=dict(size=14.5), align="left"),
|
| 124 |
cells=dict(
|
| 125 |
values=rows,
|
| 126 |
align="left",
|
| 127 |
+
font=dict(size=14.5),
|
| 128 |
height=30,
|
| 129 |
),
|
| 130 |
columnwidth=column_widths,
|
|
|
|
| 206 |
"gpt-4o-2024-05-13",
|
| 207 |
"llava-hf/llava-v1.6-vicuna-7b-hf"
|
| 208 |
]
|
| 209 |
+
perspectives = ["Safety", "Fairness", "Hallucination", "Privacy", "Adv", "OOD"]
|
| 210 |
main_scores_t2i = {}
|
| 211 |
main_scores_i2t = {}
|
| 212 |
sub_scores_t2i = {}
|